IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 1, March 2024, pp. 619~630

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i1.pp619-630  619

Application of machine learning in chemical engineering:

outlook and perspectives

Ashraf Al Sharah1, Hamza Abu Owida2, Feras Alnaimat2, Suhaila Abuowaida3

1
Department of Electrical Engineering, College of Engineering Technology, Al-Balqa Applied University, Amman, Jordan
2
Department of Medical Engineering, Faculty of Engineering, Al-Ahliyya Amman University, Amman, Jordan
3
Department of Computer Science, Prince Hussein Bin Abdullah Faculty of Information Technology, Alal-Bayt University,
Mafraq, Jordan

Article Info ABSTRACT

Article history: Chemical engineers' formulation, development, and stance processes all
heavily rely on models. The physical and economic consequences of these
Received Apr 9, 2023 decisions can have disastrous effects. Attempts to employ a hybrid form of
Revised Sep 23, 2023 artificial intelligence for modeling in various disciplines. However, they fell
Accepted Nov 6, 2023 short of expectations. Due to a rise in the amount of data and computational
resources during the previous five years. A lot of recent work has gone into
developing new data sources, indexes, chemical interface designs, and
Keywords: machine learning algorithms in an effort to facilitate the adoption of these
techniques in the research community. However, there are some important
Applications downsides to machine learning gains. The most promising uses for machine
Chemical engineering learning are in time-critical tasks like real-time optimization and planning that
Machine learning require extreme precision and can build on models that can self-learn to
Models recognize patterns, draw conclusions from data, and become more intelligent
Optimization over time. Due to their limited exposure to computer science and data analysis,
the majority of chemical engineers are potentially vulnerable to the
development of artificial intelligence. But in the not-too-distant future,
chemical engineers' modeling toolbox will include a reliable machine learning
component.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Ashraf Al Sharah
Department of Electrical Engineering, College of Engineering Technology, Al-Balqa Applied University
Amman 001962, Jordan
Email: aalsharah@bau.edu.jo

1. INTRODUCTION
For the past 130 years, mathematical modelling has been a crucial tool in chemical engineering,
allowing engineers to quickly identify and design chemical processes [1], [2]. Keeping up with the ever-
changing demands of today's world is harder than ever. No matter if you're trying to discover and synthesize
active pharmaceutical ingredients to treat new diseases or increase process efficiency to conform to stricter
environmental legislation, the ability to predict the outcomes of certain events is essential. The efficiency of a
chemical interaction, the choice of a reactor, and the regulation of a heat source are all examples. Theories that
have been refined over time of several centuries, one can make predictions [3]–[5]. Consequently, for
reasonable processes, several of these models can somehow be modeled mathematically and necessitate a great
deal of supercomputing capacity to solve numerical results. Because of this limitation, most engineers resort
to more elementary models when attempting to explain the world around them. Prandtl's boundary layer model
[6] is a notable example of a model from the past that is still useful today. Scientists and engineers in the field
of computational chemistry often compromise precision for the sake of efficiency. It is because of this openness

Journal homepage: http://ijai.iaescore.com

620  ISSN: 2252-8938

that concentration structural functionalism has become so widely used in place of more advanced theoretical
models.
On the other hand, there are many scenarios where greater precision is preferred. Scientists and
engineers in the field of chemical technology now has access to a wealth of data collected over many years of
modeling, simulation, and experimentation, giving researchers an additional modeling tool in the form of the
ability to draw on prior experience to make predictions. To put it simply, machine learning models are a subset
of statistical and mathematical models that can "learn" from data by watching their environment and
"uncovering" relationship between the data without the usage of predetermined rules. Figure 1 shows the block
diagram explains the working of machine learning algorithm.

Figure 1. The block diagram explains the working of machine learning algorithm

Machine learning is an AI subfield (AI). Definition of artificial intelligence (AI): the computer's
ability to mimic human thought and behavior in certain situations. The study of these topics is not a cutting-
edge endeavour. The term "artificial intelligence" was first used in 1956 at a summer meeting at Dartmouth
College, USA, attended by mathematicians interested in developing smarter robots.
Efforts to implement AI in the field of chemical engineering didn't begin to gain traction until well
after the year 2000 [7]. Rule-based expert systems, one of the earliest and most basic types of AI, saw increased
use in the field during the 1980s. The field of machine learning had already begun to expand by that point, but
with a few notable exceptions, the chemical engineering community lagged behind by roughly ten years.
During the 1990s, there was a rapid uptick in articles on artificial intelligence progress in the field of chemical
engineering due to the widespread use of cluster analysis, optimization computation, and, most effectively,
artificial neural networks (ANNs) (ANNs). However, this fad did not last, and experts point to the absence of
potent modelling and the challenging charge of developing the algorithms as potential reasons. In the past ten
years, advancements have been made in deep learning, a branch of machine learning that builds ANNs to
simulate the human brain. While ANNs did see increased adoption in the 1990s, the advent of the deep learning
era made it possible to develop multi-layered neural networks, or "deep neural networks," which had previously
been computationally prohibitive.
Chemical engineers were set off by these innovations, as evidenced by the meteoric rise in related
research papers. The question of whether or not artificial intelligence techniques have advanced to the point
where they can be considered a typical chemical engineering tool [7]–[9]. In this review article, we will begin
by providing an overview of the three major links that currently exist in machine learning as it relates to
chemical engineering. We will examine the benefits and drawbacks of machine learning in chemical
engineering, presenting a set of hypotheses explaining why machine learning might be useful in the field of
chemical engineering, will continue to either be a "hot" topic or become obsolete in the near future.

2. MACHINE LEARNING; DATA, REPRESENTATIONS, AND MODELS

Machine learning relies on three main components: data, representations, and models. In order to train
a machine learning model, it is necessary to first collect the necessary training data. We'll get into how the data
ends up being the machine learning process's biggest flaw in its own right later on. All sorts of data, from
experiments to theoretical findings to simulation results, can be used to train a model. However, due to the high
cost of data collection in massive quantities, it is common practice to employ big data, which involve the use
of large databases culled from a variety of existing sources. Because of the high cost of conducting actual
experiments, to get such enormous amounts of data, experts use quick algorithms or information extraction of
trademarks and publications. Due to researchers' growing comfort with using digital tools, there are now many
free and paid databases available to them [10]–[12]. In order to compare the performance of different machine
learning models, several benchmark datasets have been created. Standard reference materials for quantum
chemical properties include QM9 and Alchemy [13], while standard reference materials for solubility include
estimated SOLubility (ESOL) [14], and FreeSolv [15]. Before incorporating a dataset into a model based on
machine learning, there are a number of checks and checks and balances that need to be carried out to guarantee

Int J Artif Intell, Vol. 13, No. 1, March 2024: 619-630

Int J Artif Intell ISSN: 2252-8938  621

adequate system performance. Data curation refers to the process of monitoring and maintaining data quality
at every stage, from creation to archiving. When it comes to how they utilize data, machine learning and deep
learning approaches differ significantly from more conventional forms of modeling. To begin with, ANNs are
capable of self-learning and training, but this process requires a large amount of data. Therefore, enormous
amounts of data points are typical in training datasets. Second, instead of dividing the dataset in half, it is
divided in three: the training set, the validation set, and the test set. In contrast to training data, verification data
is kept separately used to objectively assess the training phase model's accuracy. The test set is the main
indicator of model quality because it evaluates the final model fit with unseen data [9], [10], [16].
A machine learning method's representation of data in the model is also essential. Even when the data
is already numeric, the model's performance can be greatly affected by the variables or features chosen to
comprise the input. The process of feature selection has been the subject of research in a number of published
works [9], [17]. Time and money could be saved by reducing the number of features used in training if the
model's accuracy is not compromised. Deep learning methods weigh feature selection less. Thus, convolution
layers of basic process variables [18]–[20]. Representing non-numerical data such molecules and reactions is
much harder.
Molecules and/or chemical reactions are frequently involved in chemical engineering tasks. Until
reliable numerical representations of these datasets are established, they cannot be used. Common methods of
representing molecular structure in software include line-based identifiers like the simplified molecular-input
lineentry system (SMILES) or the international union of pure and applied chemistry (IUPAC) international
chemical identifiers (InChIs) [21], [22] or as three-dimensional (3D) coordinates.
Recently, a molecular string representation tailored to machine learning applications called self-
referencing embedded strings (SELFIES) [23] has been developed. As input to a deep neural network or other
machine learning model, the molecular data is transformed into a feature vector or tensor. The molecular
weight, dipole moment, and dielectric constant are examples of good molecular descriptors that can be used to
represent molecules [24], [25].
The 3D geometry of the molecule can also be used as a starting point for generating a feature vector.
Examples of geometry-based representations include coulomb matrices [26], bond bags [27], and distance,
angle, and dihedral histograms [28]. Nonetheless, many uses don't have access to 3D coordinates or calculated
properties. When this is the case, so-called topology-based representations can be built from a molecular graph
[29].
The only form of identification possible in topological representations is a line label. Natural language
processing (NLP) techniques can be used by some encoders to directly convert the line-based identifier into a
representation [30], [31]. This is accomplished by adding some simple characteristics to the linear combination,
such as particles and interactions, and then passing data back and forward between them in an incremental way
[32].
Some of the earliest molecular representations used in machine learning were circular fingerprints
[33], [34] constructed using the Morgan algorithm [35], including the extended-connectivity fingerprint [36].
Due to the fact that they remain unchanged throughout the machine learning model's training process, these
fingerprints are known as fixed molecular representations. They continue to be widely used in drug design
because of their speed and accuracy in predicting physical, chemical, and biological properties of potential new
drugs [37]. Given that the definition of a deep neural network assumes it will learn the important features, a
fixed representation vector's use as an input layer seems at odds with this assumption [38], [39], so the focus
has shifted from manually engineering the feature vector to learning how to represent a molecule [40]. To aid
in this foresight, a model is constructed that incorporates learned molecular representations. Through training,
a molecular representation is constructed and refined, beginning with elementary properties of molecules like
heavy atoms, bond types, and ring features.
This choice also hints at the fact that there are different molecular representations suitable for different
kinds of prediction jobs. Gilmer et al. [41] summarize the message-passing neural network framework, which
is used to characterize a variety of learned topology-based representations [40]. An important feature of
message-passing neural networks is the weighted transfer of atomic and bond information across the molecular
graph. Even though many representations exist, their levels of complexity range widely, and no single
representation has been developed to work for all types of molecular properties [42], [43]. When compared to
molecules, chemical reactions are much more complex. Reactions can be identified using line-based molecular
identifiers like reaction SMILES [44] and reaction InChI (RInChI) [45], while reaction mechanisms can be
determined using the SMIRKS [44] system. Like molecular interactions, chemical reactions can be vectorised
for incorporation into machine learning models. For the quickest and easiest results, begin with the molecular
descriptors (such as fingerprints) of the reagents and add, subtract, or concatenate [46], [47]. An alternate
approach is to memorize a representation of the reaction that is built around the atoms and bonds that are
actually involved in the process. A neural machine translator can be used to translate the names of organic
reaction products that have been stored as text (typically InChI) [45], [48].
Application of machine learning in chemical engineering: outlook and perspectives (Ashraf Al Sharah)
622  ISSN: 2252-8938

The final step in any machine learning procedure is a modeling strategy. There are many different
kinds of machine learning models. While regression and classification are two of the most common uses for
models, there are other classification schemes available, such as those based on various forms of machine
learning (unsupervised, supervised, active, or transfer learning) [49], [50]. As commonly understood machine
learning can be thought of as any method that implicitly models correlations within datasets. Accordingly,
many of the techniques we now refer to as machine learning were in fact employed for some time before the
term was coined to describe them. Two such methods are principal component analysis (PCA) and the Gaussian
mixture model, both of which emerged in the late 1800s [51] and early 1900s [52]. These two use-cases are
now formally represented as unsupervised machine learning algorithms. Many unsupervised clustering
methods exist, including t-distributed stochastic neighbor embedding (t-SNE) [53] and density-based spatial
clustering of applications with noise (DBSCAN) [54]. The goal of unsupervised learning is to train a model
without giving the algorithm any "solutions" or "labels" to work with as it discovers patterns on its own.
Unsupervised learning techniques have found a number of uses in chemical engineering. Palkovits R and S
Palkovits [55] used the k-means algorithm [56] to classify groups of catalysts, and t-SNE was used to visualize
the resulting high-dimensional representations. t-SNE has been implemented in several different fields outside
of catalysis, including chemical process fault diagnosis [57] and reaction-state prediction [58]. Principal
component analysis (PCA) is another dimensionality reduction algorithm commonly used by chemical
engineers to find the features that best explain the data in the training set [59]. Additional applications of PCA
include the detection of outliers [60]. Other algorithms for spotting anomalies include deep belief subspace
analysis (DBSCAN) and long short-term memory (LSTM) [61].
Supervised classification methods like decision trees (and, by extension, random forests) can be used
[62] when the dataset is labelled, meaning the correct classification of each data point is known. Alternative
supervised classification strategies include support vector machines [63]. While initially developed for
classifying data, support vector machines have since had their functionality expanded to also perform
regression. Although in principle any supervised learning method can be incorporated into an active learning
approach, it is necessary to use supervised or active learning techniques for regression problems. ANNs, in
their many forms [64], [65] are the method most commonly associated with machine learning. Feed-forward
ANNs are used for feature-based classification and regression, while convolutional neural networks are used
in image processing, and recurrent neural networks are used in natural language processing (for anomaly
detection). It is possible for a chemical engineer to come across ANNs [66], [67] support vector machines [63],
or kernel ridge regression [68] used to predict the properties of the representations, as well as convolutional
neural networks used to represent molecules [69]. Many applications in catalysis [70], chemical process control
[71], and chemical process optimization [72] have used ANNs as a black-box modeling tool. For example, k-
nearest neighbors has been implemented in applications such as chemical process monitoring [73] and catalyst
clustering [74] because it is effective at classifying data when the labels are already known. Figure 2 shows
three main connections between machine learning and chemical engineering.

Figure 2. Three main connections between machine learning and chemical engineering,
opportunities and benefits

Recent advances have provided ways to counteract some of the most significant criticisms of machine
learning methods, while the methods' much strength open up a wide range of potential applications. Nearly all
trained machine learning methods have exceptionally fast execution speeds, making them ideal for use cases
where precision and throughput within strict system constraints are paramount. High-frequency, real-time
optimization and feed- forward control of processes are two examples of such uses [75], [76]. Although detailed
fundamental models are usually not fast enough to avoid computational delays, empirical models are often
inaccurate for these applications. By being trained on the same fundamental model, machine learning models

Int J Artif Intell, Vol. 13, No. 1, March 2024: 619-630

Int J Artif Intell ISSN: 2252-8938  623

can rival the accuracy of empirical models while only requiring a fraction of the processing power. Here, a
model is trained using high-level data to estimate the discrepancy between the observed result and the correct
one [77], [78]. It has been shown that unsupervised algorithms are superior to supervised ones for spotting
anomalies in real-time data, which is useful in process control applications. Better digital twins and improved
control could lead to more productive chemical processes if faster, more precise predictions were combined
with trustworthy industrial data [79], [80].
The same holds true for multiscale modeling approaches, where phenomena are modeled at multiple
scales. This leads to an extremely intricate and tightly coupled system of equations. Machine learning's
potential in such contexts is highly context-dependent. Machine learning is not recommended if the goal is to
gain fundamental insights into the lower scale phenomena because of its black-box nature. With the
incorporation of the lower dimensions into the strategy to create a more appropriate prediction for grander scale
processes [81]–[83], machine learning has the potential to substitute the slow core frameworks for the smaller
scales without harming the comprehensibility of the larger scale behaviours.
One last chance can be found in fixing machine learning's most serious shortcoming: its inability to
be understood by humans. As it turns out, chemical engineering problems aren't the only ones where the
problem of interpretable machine learning systems arises [83]–[85]. In the area of catalysis, researchers have
tried to put numbers on just what it is that machine learning models pick up on [86]. Despite this effort, no
direct interpretation of the model results is provided. The flow process that can be used to describe how and
why a certain end result is reached, with a good result from the model, like the correct product from a chemical
reaction predictor, should not be blindly accepted without first investigating the model's assumptions. It is
helpful to get a sense of the model's confidence in its own decisions by first quantifying the individual
prediction uncertainties and then moving on to the model's output [87]–[89]. Ensemble modeling is one easy
method for doing so. For decades, meteorologists have relied on this method, which can be used in tandem
with practically any model [90], [91]. Several algorithms have also been developed to ascertain the extent to
which particular input features affect the output, or to ascertain the training points used by the model to produce
a particular output [91], [92]. Given that human interpretation of the molecular fingerprints used as input to
very complex recurrent neural networks is already a formidable challenge, the task of interpreting the output
of such a model becomes even more so. In risk management, the as low as reasonably practicable (ALARP)
principle is widely applied. A similar "as simple as reasonably possible" principle could be proposed for
machine learning models to ensure they are as interpretive as possible [83], [93].
The fields of chemical chemistry and chemical engineering have seen a rise in the popularity of
machine learning techniques because they can find trends in data that human researchers miss. Contrary to
mathematical model, which are based on clear physical formulas, machine learning models are not intended to
tackle a particular issue (resulting from discovered patterns). In contrast, physical models can be thought of as
representations of the real world. This indicates that solving classification problems does not necessitate the
programming of a single explicitly defined decision function. Accordingly, solving regression problems does
not necessitate deriving or parameterizing specific model equations [83], [91]. These advantages allow for
efficient upscaling to large systems and datasets without requiring a large amount of computing power.
Predicting quantum chemical properties using machine learning has seen a recent uptick in interest.
Calculating the characteristics of an individual atom using traditional ab initio approaches might take
many up to several hours. Machine learning algorithm that has been properly trained can make precise
predictions in a nanosecond. No doubt, other fast methods that can make accurate predictions have already
been developed; however, in comparison to machine learning models, the application range for these methods
is quite limited [94]. Machine learning's biggest shortcoming is that it can't extrapolate, but its usefulness can
be greatly increased by simply analyzing more data. Using active learning [95], we can increase the scope with
comparatively little additional information. This works wonderfully in situations where labels are costly to
obtain (such as in quantum chemical calculations [96] or chemical experiments [97]). The amount of new data
needed to increase the range can also be decreased with the help of active learning. In addition, preexisting
machine learning models like ChemProp [98] and SchNet [99] can be utilized without the need for training or
education. The accessibility of machine learning has increased thanks to the development of scikit-learn [100]
and TensorFlow [101], as well as Keras [102] and PyTorch [103]. Because of these frameworks, deep learning
model training is limited to a manageable number of lines of code. Since these libraries and frameworks already
exist, scientists can focus on the implications of their work in the real world, rather than wasting time on
developing abstract simulations [83]. Figure 3 (see an Appendix) shows the use of machine learning for
predictive modeling in chemical engineering: opportunities, and benefits.

3. CHALLENGES AND DRAWBACKS

The scientific community benefits and faces risks from the widespread availability of machine
learning models. Machine learning has the potential to benefit anyone with some programming experience, but
Application of machine learning in chemical engineering: outlook and perspectives (Ashraf Al Sharah)
624  ISSN: 2252-8938

it also leaves it vulnerable to abuse by those who don't have a thorough understanding of the underlying
algorithms [91], [104]. Parameter and hyper parameter space in contemporary machine learning algorithms is
vast. Even for the most experienced professionals, machine learning is largely a trial-and-error process. Some
people view machine learning as a type of contemporary alchemy because researchers are frequently unwilling
to actually identify how one system operates whereas other doesn't really [105]. While the fields of chemistry
and chemical engineering are not as vulnerable to a reproducibility crisis as the social sciences [106], growing
scepticism in the community may be a result of the increasingly irreproducible use of machine learning. After
reaching the peak of exaggerated expectations [107] in Gartner's hype cycle [108], machine learning and deep
learning now face the prospect of entering a period of disillusionment where interest is all but dead. Not
properly understanding the results of algorithmic analyses is just as risky [83]. It can be very difficult to
understand the reasoning behind an algorithm's output because of how opaque the algorithms themselves are.
However, sometimes a model will get the right answer for the wrong reasons [87]. This means that researchers
employing machine learning must keep in mind a fundamental statistical principle: Relationships, not
causation, are what matter most. The misuse of a machine learning model occurs when it is applied to a setting
for which it was not intended.
Limitations on applicability result from the information that was used during training. Researchers
need to make sure they're covering all of the bases by testing over a wide enough range. A user should be aware
that the model's performance will suffer if the points are outside the range [91], [109]. There are open-source
programs that employ clustering algorithms for evaluating the data's credibility and its applicability in different
settings [110]. The growing skill gap in machine learning (ML) threatens the widespread implementation of
ML in chemical engineering study. When applying computer and data science to chemistry and chemical
engineering, expertise in both the tool and the process at hand is required. Therefore, it is possible that
elementary training in employing machine learning algorithms will be insufficient in the near future [83].
Instead, it will become increasingly important for undergraduates to have a firm grasp of AI and
statistical methods in the field of chemical engineering. However, there needs to be more collaboration between
IT professionals and other specialists. Researchers who aren't adequately prepared to use computational tools
may make mistakes, and computer and data scientists who don't have enough background in the field may have
to settle for subpar results. If experts in machine learning and chemistry worked together more often, this period
of disillusionment could have been avoided [83], [111], [112].
It's a big problem that many forms of machine learning aren't particularly open to inspection. When a
certain set of parameters is provided, all of the procedures yield the same output. A model's statistical
performance on a test dataset can provide inferences about the quality of the generated output. Analyzing the
model's hyperparameters (such as the ANN's node count) can provide insight into the relationships the model
has learned to make, but it can be a time- consuming process. Machine learning models, despite being fast and
accurate, are therefore not a good option for modeling in explanatory research [113]–[115].
This lack of interpretability adds complexity to the task of designing an efficient machine learning
model. Like any other model, the best machine learning model will have some degree of optimal overfitting
and under fitting. The risk of overfitting is typically much higher for machine learning models than the risk of
under fitting, with both factors depending on the quality and quantity of the training data and the complexity
of the model. There is no way to avoid an over fitted model when attempting to use a polynomial of very high
order to fit a (noisy) linear dataset. In deep learning, overfitting typically manifests as overtraining. This means
that the model will store away meaningless blips of data rather than actual patterns. Comparing the model's
performance on the training data to that on the validation and test data is a sure-fire way to spot overtraining
[83], [116]–[118].
In the event that training performance greatly exceeds validation performance, the model may have
been overtrained. Estimating the complete number of training iterations can be difficult. Similarly to other
optimization problems, machine learning models require stopping criteria to prevent overfitting [119]–[121].
In most cases, machine learning models can perform quite accurately on the training dataset; rather,
the difficulty lies in performing well on data that was not used to train the model. As a result, the validation
dataset, which contains data that has not been used to train the model, should be used as the stopping criterion.
It is standard practice in conventional modeling methods to use a separate, independent dataset called the test
dataset to rigorously test the optimized dataset [113]–[115].
One last, often crucial, flaw in machine learning techniques is the data itself. According to the
"garbage in-garbage out' (GIGO) principle [122], a network will produce garbage results if the dataset contains
too many systematic errors. There are some mistakes that can be easily identified as to their cause or origin,
while others, once they have been made, can be extremely challenging to track down. It's possible for outliers
to appear in any statistical technique. When compared to a large dataset, outliers have a greater impact on a
model trained on a small dataset. Because of this, machine learning benefits from both high-quality data and
large amounts of data. Manually removing these outliers from the dataset is one approach to dealing with

Int J Artif Intell, Vol. 13, No. 1, March 2024: 619-630

Int J Artif Intell ISSN: 2252-8938  625

systematic errors; alternatively, anomaly detection algorithms like principal component analysis, t-SNE,
DBSCAN, and recurrent neural networks (LSTM networks) can be used. Anomaly detection methods based
on self-learning unsupervised neural networks have recently been developed [83], [91], [114], [115].
Many years of modeling, simulating, and experimenting have resulted in a massive amount of data
for the chemical engineering community, but this information is typically locked away in private archives at
universities and private companies. It is possible that even when data is readily available, such as from an
internal database, the data is not optimal for machine learning. Text-mining methods used to extract
information from scholarly articles and patents yield identical results [123]. For one thing, only positive results
from experiments are typically reported, while negative results are often ignored. As a result of the engineer's
superior wisdom and training, the chemical process is not subjected to absurd experimental or operating
conditions. Machine learning algorithms, however, are unaware of these limitations, and excluding "trivial"
data like this could have disastrous results. Figure 4 shows the use of machine learning for predictive modeling
in chemical engineering: challenges and drawbacks.

Figure 4. Use of machine learning for predictive modeling in chemical engineering: challenges and
drawbacks

4. CONCLUSIONS AND OUTLOOK

In the last decade, machine learning has become an increasingly useful tool in the arsenal of a chemical
engineer. There is a significant and expanding interest in machine learning among chemical engineers. The
rapid processing times, adaptability, and user-friendliness of machine learning-based applications are all
contributing to their growing popularity.
The flip side of machine learning's rising popularity is the chance that it will be misused, or that
chemical engineers will incorrectly interpret black-box results, leading to widespread scepticism. When people
don't trust one another, it can cause problems on the playing field. Each of the three presented ideas has the
potential to make machine learning models more credible and transform them into a more useful and
trustworthy modeling strategy.
First off, everyone in the community needs to be able to freely and unrestrictedly access the
community's data and models. When researchers have access to high-quality data and open- source models,
they are encouraged to use machine learning as a tool because it allows them to focus more on their research
topic and less on the programming and data collection required for it. The second point, the creation of
understandable models, is intrinsically linked to the first. New models for chemical applications often take
their cue from preexisting algorithms, as machine learning is already well-established in other research fields.
Therefore, instead of keeping things as black boxes, the field would benefit most from studying the
reasons behind why a certain output is generated from a given input. As a final piece of guidance, consider
using some of your resources to acquire a deep understanding of algorithmic theory. Understanding the
computer science that lies behind the graphical interface is essential for any modeler, even though chemical
engineers typically have very strong mathematical and modeling skills. This is due to the fact that the graphical
user interface is employed to symbolize the information being modeled. In addition, this should make it
possible to specify the model's usable domain, which is crucial for figuring out whether the model is
interpolating or extrapolating. Without a doubt, the most crucial consideration is the last one. Machine learning

Application of machine learning in chemical engineering: outlook and perspectives (Ashraf Al Sharah)
626  ISSN: 2252-8938

models should be trusted, but this is only possible if their credibility is monitored for any instances in which
the model is being applied to data that was not included in its training set.

APPENDIX

Figure 3. Use of machine learning for predictive modeling in chemical engineering: opportunities,
and benefits

REFERENCES
[1] C. Boyadjiev, “Theoretical chemical engineering: Modeling and simulation,” Theoretical Chemical Engineering: Modeling and
Simulation, pp. 1–594, 2010, doi: 10.1007/978-3-642-10778-8.
[2] M. Haghighatlari and J. Hachmann, “Advances of machine learning in molecular modeling and simulation,” Current Opinion in
Chemical Engineering, vol. 23, pp. 51–57, 2019, doi: 10.1016/j.coche.2019.02.009.
[3] T. G. Dobre and J. G. Marcano., Chemical engineering: Modeling, simulation and similitude. John Wiley & Sons, 2007.
[4] G. Towler and R. Sinnott, “Chemical Engineering Design: Principles, Practice and Economics of Plant and Process Design,”
Chemical Engineering Design: Principles, Practice and Economics of Plant and Process Design, pp. 1–1027, 2021, doi:
10.1016/B978-0-12-821179-3.01001-3.
[5] D. R. Baughman and Y. A. Liu, “Neural Networks in Bioprocessing and Chemical Engineering,” Academic press, 2014.
[6] O. A. Oleinik and V. N. Samokhin, “Mathematical Models in Boundary Layer Theory,” Mathematical Models in Boundary Layer
Theory, 2018, doi: 10.1201/9780203749364.
[7] T. E. Quantrille and Y. A. Liu, “Artificial Intelligence in Chemical Engineering,” 2012.
[8] L. He et al., “Applications of computational chemistry, artificial intelligence, and machine learning in aquatic chemistry research,”
Chemical Engineering Journal, vol. 426, p. 131810, Dec. 2021, doi: 10.1016/j.cej.2021.131810.
[9] L. H. Chiang, B. Braun, Z. Wang, and I. Castillo, “Towards artificial intelligence at scale in the chemical industry,” AIChE Journal,
vol. 68, no. 6, Jun. 2022, doi: 10.1002/aic.17644.
[10] J. Jiang, B. Cui, and C. Zhang, “Basics of Distributed Machine Learning,” pp. 15–55, 2022, doi: 10.1007/978-981-16-3420-8_2.
[11] N. Artrith et al., “Best practices in machine learning for chemistry,” Nature Chemistry, vol. 13, no. 6, pp. 505–508, Jun. 2021, doi:
10.1038/s41557-021-00716-z.
[12] A. Pomberger et al., “The effect of chemical representation on active machine learning towards closed-loop optimization,” Reaction
Chemistry & Engineering, vol. 7, no. 6, pp. 1368–1379, 2022, doi: 10.1039/D2RE00008C.
[13] G. Chen et al., “Alchemy: A Quantum Chemistry Dataset for Benchmarking AI Models,” 2019, [Online]. Available:
http://arxiv.org/abs/1906.09427
[14] J. S. Delaney, “ESOL: Estimating Aqueous Solubility Directly from Molecular Structure,” Journal of Chemical Information and
Computer Sciences, vol. 44, no. 3, pp. 1000–1005, May 2004, doi: 10.1021/ci034243x.
[15] D. L. Mobley and J. P. Guthrie, “FreeSolv: a database of experimental and calculated hydration free energies, with input files,”
Journal of Computer-Aided Molecular Design, vol. 28, no. 7, pp. 711–720, Jul. 2014, doi: 10.1007/s10822-014-9747-x.
[16] E. Tomelleri, L. Belelli Marchesini, A. Yaroslavtsev, S. Asgharinia, and R. Valentini, “Toward a Unified TreeTalker Data Curation
Process,” Forests, vol. 13, no. 6, p. 855, May 2022, doi: 10.3390/f13060855.
[17] D. M. Probst et al., “Evaluating Optimization Strategies for Engine Simulations Using Machine Learning Emulators,” Journal of
Engineering for Gas Turbines and Power, vol. 141, no. 9, Sep. 2019, doi: 10.1115/1.4043964.
[18] N. Buduma, “Fundamentals of Deep Learning,” CEUR Workshop Proceedings, vol. 1542, pp. 33–36, 2015.
[19] N. Thuerey, P. Holl, M. Mueller, P. Schnell, F. Trost, and K. Um, “Physics-based Deep Learning,” 2021, [Online]. Available:
http://arxiv.org/abs/2109.05237
[20] W. Q. Yan, “Computational Methods for Deep Learning,” p. 134, 2021, [Online]. Available: http://link.springer.com/10.1007/978-
3-030-61081-4
[21] D. Weininger, “SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules,” Journal
of Chemical Information and Computer Sciences, vol. 28, no. 1, pp. 31–36, Feb. 1988, doi: 10.1021/ci00057a005.
[22] S. Heller, A. McNaught, S. Stein, D. Tchekhovskoi, and I. Pletnev, “InChI - the worldwide chemical structure identifier standard,”
Journal of Cheminformatics, vol. 5, no. 1, p. 7, Dec. 2013, doi: 10.1186/1758-2946-5-7.

Int J Artif Intell, Vol. 13, No. 1, March 2024: 619-630

Int J Artif Intell ISSN: 2252-8938  627

[23] M. Krenn, F. Häse, A. Nigam, P. Friederich, and A. Aspuru-Guzik, “Self-referencing embedded strings (SELFIES): A 100% robust
molecular string representation,” Machine Learning: Science and Technology, vol. 1, no. 4, p. 045024, Dec. 2020, doi:
10.1088/2632-2153/aba947.
[24] Y. Amar, A. M. Schweidtmann, P. Deutsch, L. Cao, and A. Lapkin, “Machine learning and molecular descriptors enable rational
solvent selection in asymmetric catalysis,” Chemical Science, vol. 10, no. 27, pp. 6697–6706, 2019, doi: 10.1039/C9SC01844A.
[25] K. K. Yalamanchi, M. Monge-Palacios, V. C. O. van Oudenhoven, X. Gao, and S. M. Sarathy, “Data Science Approach to Estimate
Enthalpy of Formation of Cyclic Hydrocarbons,” The Journal of Physical Chemistry A, vol. 124, no. 31, pp. 6270–6276, Aug. 2020,
doi: 10.1021/acs.jpca.0c02785.
[26] M. Rupp, A. Tkatchenko, K.-R. Müller, and O. A. von Lilienfeld, “Fast and Accurate Modeling of Molecular Atomization Energies
with Machine Learning,” Physical Review Letters, vol. 108, no. 5, p. 058301, Jan. 2012, doi: 10.1103/PhysRevLett.108.058301.
[27] K. Hansen et al., “Machine Learning Predictions of Molecular Properties: Accurate Many-Body Potentials and Nonlocality in
Chemical Space,” The Journal of Physical Chemistry Letters, vol. 6, no. 12, pp. 2326–2331, Jun. 2015, doi:
10.1021/acs.jpclett.5b00831.
[28] F. A. Faber et al., “Prediction Errors of Molecular Machine Learning Models Lower than Hybrid DFT Error,” Journal of Chemical
Theory and Computation, vol. 13, no. 11, pp. 5255–5264, Nov. 2017, doi: 10.1021/acs.jctc.7b00577.
[29] Z. Dong, H. Lin, and J. Chen, “Computational Topology and its Applications in Geometric Design,” Recent Patents on Engineering,
vol. 16, no. 5, 2021, doi: 10.2174/1872212115666210901124742.
[30] S. Liu, M. F. Demirel, and Y. Liang, “N-gram graph: Simple unsupervised representation for graphs, with applications to
molecules,” Advances in Neural Information Processing Systems, vol. 32, pp. 8464–8476, 2019.
[31] B. Fabian et al., “Molecular representation learning with language models and domain-relevant auxiliary tasks,” 2020, [Online].
Available: http://arxiv.org/abs/2011.13230
[32] S. Kearnes, K. McCloskey, M. Berndl, V. Pande, and P. Riley, “Molecular graph convolutions: moving beyond fingerprints,”
Journal of Computer-Aided Molecular Design, vol. 30, no. 8, pp. 595–608, 2016, doi: 10.1007/s10822-016-9938-8.
[33] A. Cherkasov et al., “QSAR Modeling: Where Have You Been? Where Are You Going To?,” Journal of Medicinal Chemistry, vol.
57, no. 12, pp. 4977–5010, Jun. 2014, doi: 10.1021/jm4004285.
[34] A. Mayr et al., “Large-scale comparison of machine learning methods for drug target prediction on ChEMBL,” Chemical Science,
vol. 9, no. 24, pp. 5441–5451, 2018, doi: 10.1039/C8SC00148K.
[35] H. L. Morgan, “The Generation of a Unique Machine Description for Chemical Structures-A Technique Developed at Chemical
Abstracts Service.,” Journal of Chemical Documentation, vol. 5, no. 2, pp. 107–113, May 1965, doi: 10.1021/c160017a018.
[36] D. Rogers and M. Hahn, “Extended-Connectivity Fingerprints,” Journal of Chemical Information and Modeling, vol. 50, no. 5, pp.
742–754, May 2010, doi: 10.1021/ci100050t.
[37] L. Pattanaik and C. W. Coley, “Molecular Representation: Going Long on Fingerprints,” Chem, vol. 6, no. 6, pp. 1204–1207, Jun.
2020, doi: 10.1016/j.chempr.2020.05.002.
[38] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, May 2015, doi:
10.1038/nature14539.
[39] R. Winter, F. Montanari, F. Noé, and D.-A. Clevert, “Learning continuous and data-driven molecular descriptors by translating
equivalent chemical representations,” Chemical Science, vol. 10, no. 6, pp. 1692–1701, 2019, doi: 10.1039/C8SC04175J.
[40] K. Yang et al., “Analyzing Learned Molecular Representations for Property Prediction,” Journal of Chemical Information and
Modeling, vol. 59, no. 8, pp. 3370–3388, Aug. 2019, doi: 10.1021/acs.jcim.9b00237.
[41] J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, and G. E. Dahl, “Neural message passing for quantum chemistry,” 34th
International Conference on Machine Learning, ICML 2017, vol. 3, pp. 2053–2070, 2017.
[42] O. A. von Lilienfeld, “First principles view on chemical compound space: Gaining rigorous atomistic control of molecular
properties,” International Journal of Quantum Chemistry, vol. 113, no. 12, pp. 1676–1689, Jun. 2013, doi: 10.1002/qua.24375.
[43] L. David, A. Thakkar, R. Mercado, and O. Engkvist, “Molecular representations in AI-driven drug discovery: a review and practical
guide,” Journal of Cheminformatics, vol. 12, no. 1, p. 56, Dec. 2020, doi: 10.1186/s13321-020-00460-5.
[44] Z. Guo et al., “Graph-based Molecular Representation Learning,” IJCAI International Joint Conference on Artificial Intelligence,
vol. 2023-August, pp. 6638–6646, 2023, doi: 10.24963/ijcai.2023/744.
[45] G. Grethe, J. M. Goodman, and C. H. Allen, “International chemical identifier for reactions (RInChI),” Journal of Cheminformatics,
vol. 5, no. 1, p. 45, Dec. 2013, doi: 10.1186/1758-2946-5-45.
[46] M. H. S. Segler and M. P. Waller, “Neural‐Symbolic Machine Learning for Retrosynthesis and Reaction Prediction,” Chemistry –
A European Journal, vol. 23, no. 25, pp. 5966–5971, May 2017, doi: 10.1002/chem.201605499.
[47] B. Sanchez-Lengeling and A. Aspuru-Guzik, “Inverse molecular design using machine learning: Generative models for matter
engineering,” Science, vol. 361, no. 6400, pp. 360–365, Jul. 2018, doi: 10.1126/science.aat2663.
[48] N. S. Eyke, W. H. Green, and K. F. Jensen, “Iterative experimental design based on active machine learning reduces the experimental
burden associated with reaction screening,” Reaction Chemistry & Engineering, vol. 5, no. 10, pp. 1963–1972, 2020, doi:
10.1039/D0RE00232A.
[49] G. Mesnil et al., “Unsupervised and Transfer Learning Challenge: a Deep Learning approach,” JMLR W& CP: Proceedings of the
Unsupervised and Transfer Learning challenge and workshop, vol. 27, pp. 97–110, 2012.
[50] S. Chakraborty, B. Uzkent, K. Ayush, K. Tanmay, E. Sheehan, and S. Ermon, “Efficient Conditional Pre-training for Transfer
Learning,” IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, vol. 2022-June, pp. 4240–
4249, 2022, doi: 10.1109/CVPRW56347.2022.00469.
[51] K. Pearson, “III. Contributions to the mathematical theory of evolution,” Proceedings of the Royal Society of London, vol. 54, no.
326–330, pp. 329–333, Dec. 1894, doi: 10.1098/rspl.1893.0079.
[52] K. Pearson, “LIII. On lines and planes of closest fit to systems of points in space,” The London, Edinburgh, and Dublin
Philosophical Magazine and Journal of Science, vol. 2, no. 11, pp. 559–572, 1901, [Online]. Available:
http://www.tandfonline.com/doi/abs/10.1080/14786440109462720
[53] V. der M. L and E. H. G, “Visualizing Data using t-SNE,” Journal of machine learning research, vol. 9, pp. 2579–2605, 2008.
[54] M. Ester, H.-P. Kriegel, J. Sander, and X. Xu, “A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases
with Noise,” Proceedings of the 2nd International Conference on Knowledge Discovery and Data Mining, pp. 226–231, 1996.
[55] R. Palkovits and S. Palkovits, “Using Artificial Intelligence To Forecast Water Oxidation Catalysts,” ACS Catalysis, vol. 9, no. 9,
pp. 8383–8387, Sep. 2019, doi: 10.1021/acscatal.9b01985.
[56] G. F. Tzortzis and A. C. Likas, “The Global Kernel $k$-Means Algorithm for Clustering in Feature Space,” IEEE Transactions on
Neural Networks, vol. 20, no. 7, pp. 1181–1194, Jul. 2009, doi: 10.1109/TNN.2009.2019722.
[57] S. Zheng and J. Zhao, “A new unsupervised data mining method based on the stacked autoencoder for chemical process fault
diagnosis,” Computers & Chemical Engineering, vol. 135, p. 106755, Apr. 2020, doi: 10.1016/j.compchemeng.2020.106755.
Application of machine learning in chemical engineering: outlook and perspectives (Ashraf Al Sharah)
628  ISSN: 2252-8938

[58] H. Gao, T. J. Struble, C. W. Coley, Y. Wang, W. H. Green, and K. F. Jensen, “Using Machine Learning To Predict Suitable
Conditions for Organic Reactions,” ACS Central Science, vol. 4, no. 11, pp. 1465–1476, Nov. 2018, doi:
10.1021/acscentsci.8b00357.
[59] F. H. Vermeire and W. H. Green, “Transfer learning for solvation free energies: From quantum chemistry to experiments,” Chemical
Engineering Journal, vol. 418, p. 129307, Aug. 2021, doi: 10.1016/j.cej.2021.129307.
[60] M. Sun et al., “Adsorption mechanism of ammonia nitrogen and phenol on lignite surface: Molecular dynamics simulations and
quantum chemical calculations,” Fuel, vol. 337, p. 127157, Apr. 2023, doi: 10.1016/j.fuel.2022.127157.
[61] X. Zhang, Y. Zou, S. Li, and S. Xu, “A weighted auto regressive LSTM based approach for chemical processes modeling,”
Neurocomputing, vol. 367, pp. 64–74, Nov. 2019, doi: 10.1016/j.neucom.2019.08.006.
[62] J. A. Richards, Remote Sensing Digital Image Analysis. Cham: Springer International Publishing, 2022. doi: 10.1007/978-3-030-
82327-6.
[63] M. Tanveer, T. Rajani, R. Rastogi, Y. H. Shao, and M. A. Ganaie, “Comprehensive review on twin support vector machines,”
Annals of Operations Research, Mar. 2022, doi: 10.1007/s10479-022-04575-w.
[64] M. Shao, Y. Lin, Q. Peng, J. Zhao, Z. Pei, and Y. Sun, “Learning graph deep autoencoder for anomaly detection in multi-attributed
networks,” Knowledge-Based Systems, vol. 260, 2023, doi: 10.1016/j.knosys.2022.110084.
[65] A. Thebelt, J. Wiebe, J. Kronqvist, C. Tsay, and R. Misener, “Maximizing information from chemical engineering data sets:
Applications to machine learning,” Chemical Engineering Science, vol. 252, p. 117469, Apr. 2022, doi: 10.1016/j.ces.2022.117469.
[66] C. A. Grambow, L. Pattanaik, and W. H. Green, “Deep Learning of Activation Energies,” Journal of Physical Chemistry Letters,
vol. 11, no. 8, pp. 2992–2997, 2020, doi: 10.1021/acs.jpclett.0c00500.
[67] A. Kurani, P. Doshi, A. Vakharia, and M. Shah, “A Comprehensive Comparative Study of Artificial Neural Network (ANN) and
Support Vector Machines (SVM) on Stock Forecasting,” Annals of Data Science, vol. 10, no. 1, pp. 183–208, 2023, doi:
10.1007/s40745-021-00344-x.
[68] A. S. Christensen, L. A. Bratholm, F. A. Faber, and O. Anatole von Lilienfeld, “FCHL revisited: Faster and more accurate quantum
machine learning,” The Journal of Chemical Physics, vol. 152, no. 4, Jan. 2020, doi: 10.1063/1.5126701.
[69] K. Ghosh et al., “Deep Learning Spectroscopy: Neural Networks for Molecular Excitation Spectra,” Advanced Science, vol. 6, no.
9, May 2019, doi: 10.1002/advs.201801367.
[70] H. Li, Z. Zhang, and Z. Liu, “Application of Artificial Neural Networks for Catalysis: A Review,” Catalysts, vol. 7, no. 10, p. 306,
Oct. 2017, doi: 10.3390/catal7100306.
[71] C. A. Grambow, Y.-P. Li, and W. H. Green, “Accurate Thermochemistry with Small Data Sets: A Bond Additivity Correction and
Transfer Learning Approach,” The Journal of Physical Chemistry A, vol. 123, no. 27, pp. 5826–5835, Jul. 2019, doi:
10.1021/acs.jpca.9b04195.
[72] A. M. Schweidtmann and A. Mitsos, “Deterministic Global Optimization with Artificial Neural Networks Embedded,” Journal of
Optimization Theory and Applications, vol. 180, no. 3, pp. 925–948, Mar. 2019, doi: 10.1007/s10957-018-1396-0.
[73] S. Yan and X. Yan, “Using Labeled Autoencoder to Supervise Neural Network Combined with k -Nearest Neighbor for Visual
Industrial Process Monitoring,” Industrial & Engineering Chemistry Research, vol. 58, no. 23, pp. 9952–9958, Jun. 2019, doi:
10.1021/acs.iecr.9b01325.
[74] A. F. Zahrt, J. J. Henle, B. T. Rose, Y. Wang, W. T. Darrow, and S. E. Denmark, “Prediction of higher-selectivity catalysts by
computer-driven workflow and machine learning,” Science, vol. 363, no. 6424, Jan. 2019, doi: 10.1126/science.aau5631.
[75] Z. Wu, D. Rincon, and P. D. Christofides, “Real-Time Adaptive Machine-Learning-Based Predictive Control of Nonlinear
Processes,” Industrial & Engineering Chemistry Research, vol. 59, no. 6, pp. 2275–2290, Feb. 2020, doi: 10.1021/acs.iecr.9b03055.
[76] Z. Zhang, Z. Wu, D. Rincon, and P. Christofides, “Real-Time Optimization and Control of Nonlinear Processes Using Machine
Learning,” Mathematics, vol. 7, no. 10, p. 890, Sep. 2019, doi: 10.3390/math7100890.
[77] A. Das, K. G. Dhal, S. Ray, and J. Gálvez, “Histogram-based fast and robust image clustering using stochastic fractal search and
morphological reconstruction,” Neural Computing and Applications, vol. 34, no. 6, pp. 4531–4554, Mar. 2022, doi:
10.1007/s00521-021-06610-6.
[78] T. Bikmukhametov and J. Jäschke, “Combining machine learning and process engineering physics towards enhanced accuracy and
explainability of data-driven models,” Computers & Chemical Engineering, vol. 138, p. 106834, Jul. 2020, doi:
10.1016/j.compchemeng.2020.106834.
[79] J. H. Cavalcanti, T. Kovács, and A. Kő, “Production System Efficiency Optimization Using Sensor Data, Machine Learning-based
Simulation and Genetic Algorithms,” Procedia CIRP, vol. 107, pp. 528–533, 2022, doi: 10.1016/j.procir.2022.05.020.
[80] E. B. Priyanka, S. Thangavel, X.-Z. Gao, and N. S. Sivakumar, “Digital twin for oil pipeline risk estimation using prognostic and
machine learning techniques,” Journal of Industrial Information Integration, vol. 26, p. 100272, Mar. 2022, doi:
10.1016/j.jii.2021.100272.
[81] M. A. Umer, K. N. Junejo, M. T. Jilani, and A. P. Mathur, “Machine learning for intrusion detection in industrial control systems:
Applications, challenges, and recommendations,” International Journal of Critical Infrastructure Protection, vol. 38, p. 100516,
Sep. 2022, doi: 10.1016/j.ijcip.2022.100516.
[82] I. Pan, L. R. Mason, and O. K. Matar, “Data-centric Engineering: integrating simulation, machine learning and statistics. Challenges
and opportunities,” Chemical Engineering Science, vol. 249, p. 117271, Feb. 2022, doi: 10.1016/j.ces.2021.117271.
[83] M. R. Dobbelaere, P. P. Plehiers, R. Van de Vijver, C. V. Stevens, and K. M. Van Geem, “Machine Learning in Chemical
Engineering: Strengths, Weaknesses, Opportunities, and Threats,” Engineering, vol. 7, no. 9, pp. 1201–1211, Sep. 2021, doi:
10.1016/j.eng.2021.03.019.
[84] D. Gunning and D. W. Aha, “DARPA’s Explainable Artificial Intelligence Program,” AI Magazine, vol. 40, no. 2, pp. 44–58, Jun.
2019, doi: 10.1609/aimag.v40i2.2850.
[85] A. Abdul, J. Vermeulen, D. Wang, B. Y. Lim, and M. Kankanhalli, “Trends and Trajectories for Explainable, Accountable and
Intelligible Systems,” in Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, Apr. 2018, pp. 1–18.
doi: 10.1145/3173574.3174156.
[86] J. A. Kammeraad, J. Goetz, E. A. Walker, A. Tewari, and P. M. Zimmerman, “What Does the Machine Learn? Knowledge
Representations of Chemical Reactivity,” Journal of Chemical Information and Modeling, vol. 60, no. 3, pp. 1290–1301, Mar.
2020, doi: 10.1021/acs.jcim.9b00721.
[87] D. P. Kovács, W. McCorkindale, and A. A. Lee, “Quantitative interpretation explains machine learning models for chemical reaction
prediction and uncovers bias,” Nature Communications, vol. 12, no. 1, p. 1695, Mar. 2021, doi: 10.1038/s41467-021-21895-w.
[88] K. Preuer, G. Klambauer, F. Rippmann, S. Hochreiter, and T. Unterthiner, “Interpretable Deep Learning in Drug Discovery,” 2019,
pp. 331–345. doi: 10.1007/978-3-030-28954-6_18.
[89] E. Begoli, T. Bhattacharya, and D. Kusnezov, “The need for uncertainty quantification in machine-assisted medical decision

Int J Artif Intell, Vol. 13, No. 1, March 2024: 619-630

Int J Artif Intell ISSN: 2252-8938  629

making,” Nature Machine Intelligence, vol. 1, no. 1, pp. 20–23, Jan. 2019, doi: 10.1038/s42256-018-0004-1.
[90] W. S. Parker, “Ensemble modeling, uncertainty and robust predictions,” WIREs Climate Change, vol. 4, no. 3, pp. 213–223, May
2013, doi: 10.1002/wcc.220.
[91] A. M. Schweidtmann et al., “Machine Learning in Chemical Engineering: A Perspective,” Chemie Ingenieur Technik, vol. 93, no.
12, pp. 2029–2039, Dec. 2021, doi: 10.1002/cite.202100083.
[92] R. Guidotti, A. Monreale, S. Ruggieri, F. Turini, F. Giannotti, and D. Pedreschi, “A Survey of Methods for Explaining Black Box
Models,” ACM Computing Surveys, vol. 51, no. 5, pp. 1–42, Sep. 2019, doi: 10.1145/3236009.
[93] L. Bruce and G. Popov, “Improving Risk Management,” Professional Safety, pp. 34–36, 2021.
[94] K. Han, A. Jamal, C. A. Grambow, Z. J. Buras, and W. H. Green, “An Extended Group Additivity Method for Polycyclic
Thermochemistry Estimation,” International Journal of Chemical Kinetics, vol. 50, no. 4, pp. 294–303, 2018, doi:
10.1002/kin.21158.
[95] J. Vandermause, Y. Xie, J. S. Lim, C. J. Owen, and B. Kozinsky, “Active learning of reactive Bayesian force fields applied to
heterogeneous catalysis dynamics of H/Pt,” Nature Communications, vol. 13, no. 1, p. 5183, Sep. 2022, doi: 10.1038/s41467-022-
32294-0.
[96] Y.-P. Li, K. Han, C. A. Grambow, and W. H. Green, “Self-Evolving Machine: A Continuously Improving Model for Molecular
Thermochemistry,” The Journal of Physical Chemistry A, vol. 123, no. 10, pp. 2142–2152, Mar. 2019, doi:
10.1021/acs.jpca.8b10789.
[97] C. Zhang, Y. Amar, L. Cao, and A. A. Lapkin, “Solvent Selection for Mitsunobu Reaction Driven by an Active Learning Surrogate
Model,” Organic Process Research & Development, vol. 24, no. 12, pp. 2864–2873, Dec. 2020, doi: 10.1021/acs.oprd.0c00376.
[98] E. B. Lenselink and P. F. W. Stouten, “Multitask machine learning models for predicting lipophilicity (logP) in the SAMPL7 challenge,”
Journal of Computer-Aided Molecular Design, vol. 35, no. 8, pp. 901–909, Aug. 2021, doi: 10.1007/s10822-021-00405-6.
[99] K. T. Schütt, P. Kessel, M. Gastegger, K. A. Nicoli, A. Tkatchenko, and K.-R. Müller, “SchNetPack: A Deep Learning Toolbox
For Atomistic Systems,” Journal of Chemical Theory and Computation, vol. 15, no. 1, pp. 448–455, Jan. 2019, doi:
10.1021/acs.jctc.8b00908.
[100] Pedregosa Fabian et al., “Scikit-learn: Machine Learning in Python,” Journal of Machine Learning Research, vol. 12, pp. 2825–
2830, 2011, [Online]. Available: http://scikit-learn.sourceforge.net.
[101] S. Ghemawat et al., “TensorFlow: A system for large-scale machine learning,” Proceedings of the 12th USENIX Symposium on
Operating Systems Design and Implementation, OSDI 2016, p. 21, 2016, doi: 10.5555/3026877.3026899.
[102] F. Chollet, “Keras: Deep Learning for humans,” Github, p. Accessed 4 June 2017, 2015, [Online]. Available:
https://github.com/keras-team/keras.
[103] A. Paszke et al., “PyTorch: An imperative style, high-performance deep learning library,” Advances in Neural Information
Processing Systems, vol. 32, 2019, [Online]. Available: https://www.semanticscholar.org/paper/PyTorch%3A-An-Imperative-
Style%2C-High-Performance-Deep-Paszke-Gross/3c8a456509e6c0805354bd40a35e3f2dbf8069b1
[104] M. Azeem, A. Haleem, and M. Javaid, “Symbiotic Relationship Between Machine Learning and Industry 4.0: A Review,” Journal
of Industrial Integration and Management, vol. 07, no. 03, pp. 401–433, Sep. 2022, doi: 10.1142/S2424862221300027.
[105] S. Milivojevic, “Artificial intelligence, illegalised mobility and lucrative alchemy of border utopia,” Criminology and Criminal
Justice, p. 174889582211238, Sep. 2022, doi: 10.1177/17488958221123855.
[106] A. Goeva, S. Stoudt, and A. Trisovic, “Toward Reproducible and Extensible Research: from Values to Action,” Harvard Data
Science Review, vol. 2, no. 4, Dec. 2020, doi: 10.1162/99608f92.1cc3d72a.
[107] J. Fenn, M. Raskino, and B. Burton, “Understanding Gartner’s Hype Cycles,” Strategic Analysis Report, no. July 2013, p. 35, 2013.
[108] S. Shin and J. Kang, “Structural features and Diffusion Patterns of Gartner Hype Cycle for Artificial Intelligence using Social
Network analysis,” Journal of Intelligence and Information Systems, vol. 28, no. 1, pp. 107–129, 2022.
[109] M. Thombre, Z. Mdoe, and J. Jäschke, “Data-driven robust optimal operation of thermal energy storage in industrial clusters,”
Processes, vol. 8, no. 2, 2020, doi: 10.3390/pr8020194.
[110] S. H. Symoens et al., “QUANTIS: Data quality assessment tool by clustering analysis,” International Journal of Chemical Kinetics,
vol. 51, no. 11, pp. 872–885, Nov. 2019, doi: 10.1002/kin.21316.
[111] M. I. Jordan and T. M. Mitchell, “Machine learning: Trends, perspectives, and prospects,” Science, vol. 349, no. 6245, pp. 255–
260, Jul. 2015, doi: 10.1126/science.aaa8415.
[112] S. Vollmer et al., “Machine learning and artificial intelligence research for patient benefit: 20 critical questions on transparency,
replicability, ethics, and effectiveness,” BMJ, p. l6927, Mar. 2020, doi: 10.1136/bmj.l6927.
[113] V. Sze, Y.-H. Chen, J. Emer, A. Suleiman, and Z. Zhang, “Hardware for machine learning: Challenges and opportunities,” in 2017
IEEE Custom Integrated Circuits Conference (CICC), Apr. 2017, pp. 1–8. doi: 10.1109/CICC.2017.7993626.
[114] A. L’Heureux, K. Grolinger, H. F. Elyamany, and M. A. M. Capretz, “Machine Learning With Big Data: Challenges and
Approaches,” IEEE Access, vol. 5, pp. 7776–7797, 2017, doi: 10.1109/ACCESS.2017.2696365.
[115] C. Rudin, C. Chen, Z. Chen, H. Huang, L. Semenova, and C. Zhong, “Interpretable machine learning: Fundamental principles and
10 grand challenges,” Statistics Surveys, vol. 16, no. none, Jan. 2022, doi: 10.1214/21-SS133.
[116] H. K. Jabbar and R. Z. Khan, “Methods to Avoid Over-Fitting and Under-Fitting in Supervised Machine Learning (Comparative
Study),” pp. 163–172, 2015, doi: 10.3850/978-981-09-5247-1_017.
[117] H. Zhang, L. Zhang, and Y. Jiang, “Overfitting and Underfitting Analysis for Deep Learning Based End-to-end Communication
Systems,” 2019 11th International Conference on Wireless Communications and Signal Processing, WCSP 2019, 2019, doi:
10.1109/WCSP.2019.8927876.
[118] J. Kolluri, V. K. Kotte, M. S. B. Phridviraj, and S. Razia, “Reducing Overfitting Problem in Machine Learning Using Novel L1/4
Regularization Method,” Proceedings of the 4th International Conference on Trends in Electronics and Informatics, ICOEI 2020,
pp. 934–938, 2020, doi: 10.1109/ICOEI48184.2020.9142992.
[119] S. Diksha and K. Neeraj, “A Review on Machine Learning Algorithms, Tasks and Applications,” International Journal of Advanced
Research in Computer Engineering & Technology (IJARCET), vol. 6, no. 10, pp. 1548–1552, 2017.
[120] A. D’Amour et al., “Underspecification Presents Challenges for Credibility in Modern Machine Learning,” 2020, [Online].
Available: http://arxiv.org/abs/2011.03395
[121] O. Gheibi, D. Weyns, and F. Quin, “Applying Machine Learning in Self-adaptive Systems,” ACM Transactions on Autonomous
and Adaptive Systems, vol. 15, no. 3, pp. 1–37, Sep. 2020, doi: 10.1145/3469440.
[122] O. R. P. Bininda-Emonds, K. E. Jones, S. A. Price, M. Cardillo, R. Grenyer, and A. Purvis, “Garbage in, Garbage out,” in
Computational biology, Springer Nature (Netherlands), 2004, pp. 267–280. doi: 10.1007/978-1-4020-2330-9_13.
[123] N. Schneider, D. M. Lowe, R. A. Sayle, M. A. Tarselli, and G. A. Landrum, “Big Data from Pharmaceutical Patents: A
Computational Analysis of Medicinal Chemists’ Bread and Butter,” Journal of Medicinal Chemistry, vol. 59, no. 9, pp. 4385–4402,
May 2016, doi: 10.1021/acs.jmedchem.6b00153.
Application of machine learning in chemical engineering: outlook and perspectives (Ashraf Al Sharah)
630  ISSN: 2252-8938

BIOGRAPHIES OF AUTHORS

Ashraf Al Sharah Ashraf Al Sharah has completed his PhD from Tennessee
State University. USA. He was a research associate at cyber vis research lab. And served as
an Assistant Professor in the Department of Computer Engineering at Al-Ahliyya Amman
University. He is serving now as an assistant professor in electrical engineering department
in Al-Balqa Applied University. His research interest includes wireless security, IoT, smart
attack, AI, and game theory. He can be contacted at email: aalsharah@bau.edu.jo.

Hamza Abu Owida Hamza Abu Owida has completed his PhD from Keele
university, UK. He was a postdoctoral Research Associate: Developing xeno-free
nanofibrous scaffold methodology for human pluripotent stem cell expansion, differentiation
and implantation towards a therapeutic product, Keele University, Institute for Science and
Technology in Medicine (ISTM), Staffordshire /UK. He is assiosiate professor in medical
engineering department in Al-Ahliyya Amman University. He has published more than 30
papers in reputed journals. He can be contacted at email: h.abuowida@ammanu.edu.jo.

Feras Alnaimat has completed his PhD from University of Birmingham,

Birmingham, UK. In 2018, he joined the department of Medical Engineering, Al-Ahliyya
Amman University, as an assistant professor. His current research interests include design of
artificial disc implant, artificial joints and bio fluid mechanics. He is one of the steering
committees of the Innovation and New Trends in Engineering, Science and Technology
Education onference. He can be contacted at email: f.alnaimat@ammanu.edu.jo.

Suhaila Abuowaida has completed her PhD from USM University, Malaysia.
received the B. Sc degrees in computer information system and M. Sc degrees in computer
science from AL al-Bayt University, Jordan, in 2012 and 2015, respectively. Her research
interests include Deep Learning and Computer vision. She can be contacted at email:
suhila@aabu.edu.jo.

Int J Artif Intell, Vol. 13, No. 1, March 2024: 619-630