Impact of Machine Learning and Internet

of Things in Agriculture: State of the Art

S.R. Juhi Reshma and Anitha S. Pillai(&)

School of Computing Sciences, Hindustan Institute of Technology and Science,

Chennai, India
jucibi@gmail.com,
anithasp@hindustanuniv.ac.in

Abstract. Majority of the population are directly or indirectly dependent on

agriculture. In this modern world, agriculture has to be supported with tech-
nology to bring the best output. There is a big revolution in agriculture from
traditional methods. Recent development in technology has a great impact on
agriculture. Evolution of Machine Learning (ML) and Internet of Things
(IoT) has helped researchers to apply these techniques in agriculture to help
farmers. This in turn helped farmers to increase the productivity, make use of
maximum land available, control pest, and so on. This paper highlights the work
done in agriculture sector using ML and IoT.

Keywords: Machine learning
IoT
SVM
ANN
Agriculture

1 Introduction

As per The World Bank study report of September 2016, “Agricultural development is
one of the most powerful tools to end extreme poverty, boost shared prosperity and
feed 9 billion people by 2050. Growth in the agriculture sector is about two to four
times more effective in raising incomes among the poorest compared to other sectors.
This is important for 78% of the world’s poor who live in rural areas and depend
largely on farming to make a living. Agriculture is also crucial to economic growth: it
accounts for one-third of gross-domestic product (GDP) and three-quarters of
employment in Sub-Saharan Africa. But agriculture-driven growth and poverty
reduction, as well as food security are at risk: A warming climate could cut crop yields
by more than 25%”. This study highlights the importance of agriculture in any country
especially the developing ones.
Agriculture is a step by step process like ploughing of soil, preparing for planting,
applying proper fertilizer, pest and disease control, to get maximum yield. For decades
agricultural science has been working to evolve technologies in farming. The major
factors affecting farming are climate change, nature of soil, population growth, pest &
diseases. Farmers should be able to produce maximum yield from the land available to
them. Studies to maximum exploit the minimum available land to bring the best yield
has been carried out. This paper is an attempt to study the various ML as well as IoT
techniques applied to the agriculture sector to predict soil moisture, weather, and other
factors affecting agriculture.
© Springer International Publishing AG 2018
A. Abraham et al. (eds.), Proceedings of the Eighth International Conference on Soft
Computing and Pattern Recognition (SoCPaR 2016), Advances in Intelligent
Systems and Computing 614, DOI 10.1007/978-3-319-60618-7_59
 Impact of Machine Learning and Internet of Things in Agriculture 603

The paper is organised as: Sect. 2, describes research activities carried out using
ML techniques in agriculture and Sect. 3 about the use of IoT in agriculture and the
paper concludes in Sect. 4.

2 Machine Learning Applied in Agriculture Sector

Well known definitions about ML states “Machine Learning is the field of study that
gives computer the ability to learn without being explicitly programmed.” [Arthur
Samuel, 1959]. “A computer program is said to learn from experience E with respect to
some task T and some performance measure P, if its performance on T, as measured by
P, improves with experience E.” [Tom Mitchell, 1979]. In a nut shell training the system
in order to do predictions. Two main classification of ML algorithms are supervised and
unsupervised. In supervised learning the system is trained on a pre-defined set of
training data which then facilitates the ability to do predictions given a new data. Where
as in unsupervised learning the system is given a bunch of data to learn on its own by
identifying patterns and relationships.
A number of models used to predict the physical properties of the soil are available.
It is very essential to learn about soil in agriculture, because at times without proper
knowledge about the physical properties and moisture present in the soil wrong
selection of crop may lead to a heavy damage. To overcome these problems a number
of researches are carried out for soil texture identification using algorithms like support
vector machines (SVM), Naïve Bayes, and Decision tree. In order to carry out studies
on Soil, a database containing all the physical properties of each soil type are created
first. Different learning algorithms are used to train the system in order to predict the
soil characteristics.
From literature review it is observed that most of the work related to Soil prediction
is carried out using supervised learning algorithms. Support Vector Machine (SVM) is
mainly used in classifying the labelled linear data set. In addition to SVM, decision tree
and Naïve Bayes are also used.
The support vector machine (SVM) is otherwise called as support vector networks,
is a supervised learning algorithm. SVM can also be used in classification of soil
patterns. Naive Bayes algorithm is based on probability. It points the maximum number
of likelihood to find the solution. In decision tree there are multiple r nodes that contain
the test cases. It is represented in tree structure, with each node containing one con-
dition. These are some of the algorithms that are used in studies related to agriculture.
Artificial Neural Networks (ANN) techniques are also used in agricultural science to
predict the soil moisture in a particular area. The data collected are trained using ANN
techniques, thus providing optimal results.
According to [5], a multilingual soil profile database was designed containing the
attributes of different soil. According to the authors Soil Data Base management (SDBm
Plus) is a powerful tool for soil characterisation and Land use Planning. It was developed
to enable detailed characterisation of any geo-referenced soil in any part of the world in
any of the languages [5]. The soil profile is grouped into different blocks. “Block 1 general
information, Block 2 Soil horizon description, Block 3: Standard analyses, Block 4:
Soluble salts and Heavy metals, Block 5: Physical data, Block 6: Water retention and
604 S.R. Juhi Reshma and A.S. Pillai

Hydraulic conductivity, Block 7: Additional analytical Variables, Block 8: Photographs,

Block 9: Metadata” [5]. Each blocks are distinct datasets. The data in the data base are
stored in coded format that helps in reducing the size of the data base.
The SDBm plus is a useful relational database tool that helps in characterising the
soil profile. This tool provides detailed knowledge about the geo-referenced soil type at
different parts of the world, it is available in different languages. It’s a user friendly
database which takes soil taxonomy as the input and provides soil properties as the
output. The SDBm plus can be used in analysing the soil quality and the soil degra-
dation of soil.
Mediterranean Land Evaluation Information System (Micro LEIS) [6], tools was
used for making decision in land evaluation. In this paper they have focused on soil
properties, since soil plays a major role in crop yield. Each soil has its own properties,
and limitations. In certain soil, few crops may not give a good yield, to solve this
problem the soil texture should be learned. The design philosophy follows a tool kit
approach, integrating database, statistics, expert system, neural networks, web, GIS
application and other information technologies [6]. Using Micro LEIS system the
inherit capacity of each soil is analysed, which in turn is used in agricultural land to
increase the productivity and to maximum utilize the available land for agriculture.
In paper [6] three different types of databases, each for soil, climate and farming are
used. These database’s are interconnected using Relational database management
system (RDBMS). The soil database contains the properties of the soil, which are
geo-referred data. The farming data are collected by conducting an interview with the
farmers, it contains data like types of crops. The database is user friendly for the users
to make use of it. A decision support system is designed to evaluate the land ecology.
Using this model the unused land can also be evaluated and made useful. In some
cases, the land might be very well fertile to be used for agriculture but are unknown.
According to [6] The land suitability/vulnerability or response variable Y is anal-
ysed as a function of the type Y = Ø (X1, X2, …, Xn) + e,
Where, Xn - corresponds to the selected land characteristics or independent vari-
ables (e.g. soil depth, clay content, organic matter, caption exchange capacity, pH,
sodium saturation, etc.),
e - Measures the residual.
Ø - the mathematical form is not known, this function can often be approximated
satisfactorily, within the experimental context, by a polynomial equation. The cali-
bration of this polynomial model can be treated statistically as a particular case of
multiple regression.
The regression coefficient (R2) by this analysis represent the inductive validation
index of the model corresponding to that accounted for by the percentage of the
observed variation [6]. The Land evaluation model uses GIS technology to generate
images. Using these images and data sets, artificial neural network techniques are used
to generate similar patterns by training the computer with data sets.
According to [9], the work is carried out on soil database provided by Australian
continent. In this paper to generate Australia-wide soil properties predictive model the
authors have taken heterogeneous data set. The data mining & knowledge discovery
from database tool is used in this work. Piecewise linear tree models were built
choosing from 19 climate variables, digital elevation model (DEM) and derived terrain
 Impact of Machine Learning and Internet of Things in Agriculture 605

attributes, four Landsat multi-spectral scanner (MSS) bands, land use and lithology
maps as predictors of topsoil and subsoil pH, organic C, % clay, and total N and P [9].
In order to estimate top soil and sub soil horizon thickness and texture class using
similar predictors, Classification trees were used.
The climate database is also generated to analyse the soil moisture and the soil
property which varies with respect to the climate. The annual temperature, precipita-
tion, highest monthly radiation, moisture index seasonality are some of the parameters
that effects the top soil pH model. Whereas for the top soil organic C, the soil moisture
is an important parameter. To estimate the % of clay in the Top soil, all the parameters
such as organic C, pH value, soil maps are linked together. The data for 20 yrs was
taken into account and regression model was used to predict the climate along with soil
texture. Here, decision tree algorithm with very limited rules was used to identify the
soil properties like top soil, clay, organic properties etc.
Paper [15] has justified the pros and cons in using support vector machines
(SVM) and Artificial neural networks in pedotransfer functions of a soil properties. This
paper deals with the comparative study of two machine learning algorithms, support
vector machines (SVM) and Artificial neural networks (ANN). To undergo this work
around 290 soil profile was collected and the data are divided in such way that, one set of
data are used for training and another set of data are used for testing. The performance of
both the algorithm were statistically measured using Root-Mean-Squared Error
(RMSE). In predicting soil water potentials the ANN performed better than SVM, the
RMSE values of SVMs were less when compared with ANNs. In low water potential the
RMSE difference between two algorithms shows very high variations and ANN per-
formed better than SVM in predicting low water potential. The R2 values shows the
performance of both the algorithm and it is clearly recorded SVM shows better accuracy
and minimal error while compared with ANN. According to the authors SVM provides
significantly reliable results when compared with ANN.
Regression techniques was used in [20], to find the soil moisture. A remote sensing
device was used to estimate the soil moisture. Soil moisture plays a major role in
understanding the hydrology and climate in a particular area. In this paper, they have
focused on river basin area. The Colorado is a river basin that was affected badly by
drought few years back. It is believed that the drought is because of the temporal
variability of the soil. This study is based on the river basin Colorado, which has
multiple vegetation around the river basin. The soil profile of low, high and densely
vegetated region are taken. Remote sensing technologies are widely used to sense the
soil moisture. SVM, ANN, MLR was used for the study. The soil moisture was
observed using Tropical Rainfall Measuring Mission in Precipitation Radar
(TRMMPR) including microwave back scatter and angle of incident measurement [20].
In interrelating the soil moisture and vegetation the SVM is able to produce better
accuracy. Results were compared and SVM model performed better for soil moisture
estimation than ANN and MLR models [20]. This method using SVM can be used as
an alternative to physical computational model in predicting soil moisture. In agri-
culture estimating soil moisture is very important to manage irrigation and to forecast
crop yield.
606 S.R. Juhi Reshma and A.S. Pillai

In [2] they have used data mining techniques to establish the relationship between
soils. Morphology of each soil type is collected from the department of soil science and
agricultural chemistry. The naïve Bayes algorithm is used to classify the soils. When
Naive Bayes Classifier is applied to the soil data set the instances are 100% classified
[2]. The Kappa statistic, Mean absolute error, Root mean squared error, Relative
absolute error are less than the remaining Classifiers, like Bayesian classifier [2].
According to the authors the time to build the Naive Bayes Classifier is less than the
remaining Classifier and they conclude that the Naive Bayes Classifier is the efficient
classification techniques classify the types of soils. From this study it is also evident
that data mining techniques can be used in soil researches.
According to [19], soil moisture of a particular area was studied by high resolution
satellites. The artificial intelligence techniques were used to increase the resolution, so
that a detailed analysis of soil moisture is obtained. The AI downscaling algorithms
generally improves the data quality and provides higher spatial data resolution in
comparison with the original Soil Moisture and Ocean Salinity (SMOS) data [19]. Here
in this paper artificial neural networks (ANN), Support Vector Machines (SVM),
Relevance Vector Machine and generalized linear model are used. The authors also
suggest that the downscaling of soil moisture can be carried out using support vector
machines (SVM), Relevance Vector Machines (RVS), and Artificial neural networks
(ANN). The RVM reduces the computational time when compared with SVM.
In [3], the study of soil taxonomy is carried out in Lower Fraser Valley, British
Columbia, Canada using different Machine learning classification algorithms like
CART, CART with bagging, Random Forest, k-nearest neighbour, nearest shrunken
Centroid, artificial neural network, multinomial logistic regression, logistic model trees,
and support vector machine. Authors have used conventional soil survey maps to gen-
erate the data about the soil taxonomy, and it is compressed into a single – component.
Using this single – component data set, random points were generated for training data
set. Environmental covariates like topographic, climatic, and vegetative indices were
taken into account to evaluate the soil taxonomy. Among the 10 machine learning
algorithms used to classify the soil taxonomy, the Random forest (RF) and support vector
machine - radial basis function (SVM-RBF) provides better accuracy in classification.
Authors have proposed an agriculture classification system that enhances the pro-
ductivity. This system enables the farmers to predict or identify agriculture properties
such as pasture, seed, leaf, growing of crops & plants [1]. Use of information and
communication technology (ICT) in agriculture facilitates increasing the productivity.
The objective of this work is to reduce the irrelevant and redundant attributes and to
bring higher accuracy in predicting the soil properties to produce better yield. A dif-
ferential evolution algorithm based feature selection [1] is proposed in this paper, the
selected data are fed into the classification tree model to design an agriculture classi-
fication system. Authors have Different classification algorithms such as function based
classifier namely library for support vector machines (LIBSVM), instance based
classifier namely K-star, and tree based namely logistic model trees (LMT) [1] were
used to compare the performance of proposed model with existing model. The data was
collected from the Weka data repository.
Drought is one among the major problem faced by people around the world.
Authors have designed a high predictive information that can help the farmers to plan
 Impact of Machine Learning and Internet of Things in Agriculture 607

and make use of available water sources by short- term forecasting. In this paper [13]
authors have designed an user friendly tool, They have created a tool that predicts
future vegetation health values at a high spatial resolution using open source tools and
data that are global in coverage [13]. They have constructed a matrix using previously
remotely sensed images from Moderate Resolution Imaging Spectroradiometer
(MODIS) datasets along with real time images. Using these images as data, a matrix is
created. The data are fed into supervised learning algorithm such as gradient boosted
machines (GBM) and deep neural networks. This model provides better accuracy in
low cloud cover area like, it still performs well in cloudiest region Sri Lanka. The tool
can be used in making decision in agriculture.
Authors [8] have used Machine Learning algorithms for solving problem in geo-
science and remote sensing. Nonparametric regression and classification purposes are
used in this paper. The main objective of this paper is to classify the surface type, it is
done using Self-organizing maps (SOMs). To design a fine gradient of classification
SOM were used to group the data of a location. Using machine learning techniques in
geoscience provides better efficiency in handling the data.
In [7], a decision support system is designed that can help farmers to manage
different actions like crop selection, crop sequence, pest and weed control for both
conventional and genetically modified crops. Here different actions such as crop
selection, crop sequence, pest and weed control are focused and each actions are coined
in different modules. In above actions soil properties plays a major role, Flora and
Fauna in the soil is responsible for the fertility and other soil properties. In this paper
microarthropods are identified and grouped using machine learning methods such as
regression tree, model tree, and linear equation. Here the authors have used machine
learning methods to model a community of microarthropods with the help of data on
chemical, biological and mechanical actions on the fields.
In [22], the authors have focused on soil properties such as organic matters,
essential plant nutrient, micronutrient that supports the growth of the crop. Supervised
learning algorithm Back propagation network (BPN) was used in this work. The data
about the soil are trained using BPN, and the factors affecting the growth of the crop is
predicted. It is carried out in three phases, Feed forward (Phase I), Back propagation of
error (Phase II), weight & bias updation (phase III). It is evident from the study that
BPN provides better accuracy in predicting the soil properties and essential plant
nutrient that are essential for the crop growth. After so many clustering trained network
BPN can behave like automated system to predict crop growth rate for given soil
properties [22].
In this paper [16], the machine learning algorithm are used to classify the agri-
cultural data. The study is carried out to analyse the performance of Support Vector
Machine (SVM), naive bayes and artificial neural network. The data set are obtained
from the agricultural researches in new zealand. The performance of the algorithm is
measured in terms of F1.

Precision  recall
Where F1 ¼ 2 
Precision þ recall
608 S.R. Juhi Reshma and A.S. Pillai

From the study it is evident that SVM shows better accuracy when compared with
naive bayes and artificial neural network. Here in this paper SVM is used in improving
classification performance of the agricultural data. The experimental results show that
the SVM is an effective method for classification of agricultural data [16].

3 Application of IoT in Agriculture – Related Work

IoT (internet of things), is a booming technology where in computing devices, digital

machines, objects, vehicles, animals or people that are provided with unique identifiers
with the ability to transfer data over a network without requiring human to human or
human to computer interaction. IoT is used in automobiles, climate control, logistics
co-ordination, crop monitoring, live stocks. In agriculture there are number of ongoing
researches, the idea of smart farming uses sensors that sense the data and the farmers
are notified via SMS or by other communication methods. IoT can be used in agri-
cultural sectors to sense the humidity, temperature, pressure in an area. There are
different sensors used for different purposes, temperature sensors, humidity sensors,
pressure sensors etc. In green house farming IoT are used to maintain the temperature.
In Paper [4], authors have used environmental sensors to design a smart farm. İt
provides farmers with a secure, easy to use, low cost, data analysis system [4].
The SMART FARM, provides a hybrid, distributed architecture. To make a real time
decision support the data are collected using sensors and then analysed. The envi-
ronmental phenomena are measured using sensors and the farm activities are recorded.
The SMART FARM can also retrieve data from external sources if an internet con-
nection is established. The external source data like weather forecasting, temperature,
etc., can be retrieved. Historical records such as sensing devices, the internet, and
imagery data [4] are also fetched to support farming, sensors are used in all activities of
farming. Using the historical data, the decision are taken for the amount of pesticides to
be used, water required, and the fertilizer to be used. The sensing devices are plotted in
the farm house to monitor the water pumping pressure, and the amount of water used
were controlled by the sensing devices. The SMART FARM make use of private cloud,
it provides infrastructure as a service (IaaS) & Platform as a service (PaaS). Here
AppScale cloud and Eucalyptus cloud are used, which are commonly used commercial
grade platform. Using the real time data along with the retrieved data from external
source, decision can be taken.
The model, to monitor soil hydraulic properties were proposed in [10]. The soil
variabilities like water storage capacity of a soil is predicted using soil samples taken
from maize cultivation land. The soil from different landscape and different depth are
taken. The soil depth and properties of soil are taken by electromagnetic survey, these
real time values are sensed using sensors and transmitted to the cloud. Regression
model is used for this study. The water table (WT) of a particular area influences the
soil moisture of that area. So the fluctuation of water level has to be monitored to
predict the soil moisture, it can help in scheduling the irrigation.
Agriculture is considered to be the backbone of developing countries like India. In
[23] factors affecting agriculture was identified and were tracked using sensors. Using
sensors the fields are monitored and irrigated. In this paper, the proposed model
 Impact of Machine Learning and Internet of Things in Agriculture 609

overcomes the limitations of General Packet Radio Service (GPRS) as it provides a

secured data transmission. The model supports 4G. The sensors are installed in dif-
ferent areas of the field, and the data are transferred using cloud gateway. Different
sensors are used to monitor the temperature, moisture, light and the rainfall. These
sensors are connected to the IoT hub, to transfer the data captured by the sensors. Using
Message Queuing Telemetry Transport (MQTT) protocols the data are further
processed.
Paper [11], has proposed a combined functioning of wireless networks along with
the remote monitoring system. The data are connected to the device by radio frequency
identification (RFID). İn agriculture IoT has tremendous uses like observing the tem-
perature level, humidity inside the green house, and the physical properties of the soil.
Multiple sensors are used to sense the temperature, humidity, and soil moisture. These
sensors can be wired or wireless. The wired sensors are directly connected to the
machine interface, whereas the wireless sensors are connected using wireless networks.
The data are transmitted from machine to machine, through wireless communication
system. The wireless communication is carried out through radio frequency. The
sensed data such as temperature, humidity, of the green house reaches the farmers
through SMS. The real time data are transferred by wireless network with the help of
machine to machine transferring protocols. Here the temperature sensors can convert
the sensed data into digital data, whereas the humidity of soil or weather cannot convert
the real time data into digital values directly, so AD converters are used. These AD
convertors can convert the analog signals to digital signal that can be feed in to MCU
(micro control unit)
In [12] they have designed an interconnected model that can monitor the crops in
different geographical area. Digital sensors are implemented in real time. The authors
suggest that the sensors to be used should not be expensive, it should run for a long
period of time and it should be able to support any technical advancement in future.
The real time values are periodically sent through SMS. The real time values of
temperature, relative humidity of air, moisture of soil, ambient light and CO2 con-
centration (only for green houses) [12] are sensed and collected in a database. Decision
support system is used to analyse the data and to alert the farmers in order to take
proper decision to control the pest, irrigation of the field etc. The data captured by the
sensors nodes are transferred to the micro controller unit to cluster all the sensor nodes
together. According to [12] it is a real time application implemented in many parts of
china
In [14], an idea of smart villages is proposed. Here multiple sensors and other
devices are used to monitor the activities in a place. The data generated by the sensors
are transferred through internet. In smart village, sensors are used in building, weather,
irrigation, farming, dairy products, health care etc. They are used to predict the smoke,
to manage the electricity use, and to fill the tank automatically when the water level
drops down. Many villages in India are dependent on agriculture and the weather
condition plays a major role in agriculture. Predicting weather is very essential in
sowing, harvesting, irrigation etc. Sensors are used to manage the available water
resource for the irrigation and to monitor the crop yield. In smart village sensors play
an important role to enhance the quality of life.
610 S.R. Juhi Reshma and A.S. Pillai

In [25], plant diseases are focused. Plant diseases has to be taken care, as it may
spoil the yield of the crop and also may damage the crop. The proposed model has four
category network sensor node, acquisition of monitoring information, sensor and
wireless communication technology, intelligent data processing [25]. The sensor nodes
are kept in the farm to collect data. The RFID technology was used to transmit the data.
The data can be viewed by the administrators, experts, and ordinary visitors [25] from
their phones, tablets or computers. The data that are collected by the sensors are put
together to create an expert database. Apriori algorithm and fuzzy reasoning was used
to detect the plant diseases with the help of experts. Different plant disease are inte-
grated in the database, by predicting the plant disease in early stage can increase the
yield of a crop and prevent damage of crop.
A new model based on Ontology, AgOnt (agricultural ontology) is proposed in [21]
and implemented for agricultural products. The ontology model is used in agricultural
products to establish relationships between them. The Authors used AgOnt to solve
sematic problems in agriculture, by focusing only on agricultural products and by
ignoring processing of agricultural products. Two kinds of relationships are established
using AgOnt,
(i) The product & its properties [21]
(ii) The products & its source [21]
A well-established AgOnt can interconnect the products and define the life cycle of
each product [21]. By developing such a model the history of an agriculture product
like type of seed, weather condition and environmental condition under which crop was
grown, can be known. The environmental condition of a crop is sensed using IoT
sensors.
According to [17], IOT can be used to interconnect the real time objects that can
provide real time data anywhere using internet of wired connections. This paper has
made use of sensors to find out the soil moisture. There are three layers in this model,
physical layer that contains physical sensors system, virtual layers that comprises
virtual sensor cloud, and the administration layer which contain administration cloud.
Each sensor hub contains multiple sensors that can sense different attributes. The
sensors are kept in the corners of the farm land to avoid damage to the sensors. The
sensors are kept at the soil so that it can sense the data and transfer the real time data.
The data can be transferred using internet or the physical nodes are connected to the
nearby receiver.
In [24], the objects are made smarter, a new term “Intelligation” is proposed by the
author. One of the major problem faced by Indian farmers is proper irrigation. The
authors suggest a model for remotely controlling the irrigation. In areas where internet
connection is not available the farmers can receive updated information through SMS.
The model helps to monitor the field so that sufficient amount of water is used. If excess
water is stagnated in farm it may lead to crop damage. Here an automated irrigation is
proposed. There are multiple sensors used to monitor the soil moisture to predict the
amount of water required for the field. The weather condition is also considered, to
avoid excess water in the field. Here in smart irrigation and the pumps are switched on
 Impact of Machine Learning and Internet of Things in Agriculture 611

for irrigation automatically, when the sufficient amount of water is drained into the
field, the pumps are tuned off automatically. This system brings a great relief to the
farmers, as it is user friendly.
In this paper [18], authors have proposed a smart irrigation model based on soil
moisture and the climatic conditions. Different sensors are installed in the field to
gather data. To predict the soil moisture sensors are deployed in the soil at different
corners of the field. Based on the data, decision are taken for the upcoming weeks, the
amount of water needed to irrigate the filed are noted. A model called Smart Irrigation
Decision Support System (SIDSS) is proposed to evaluate the amount of water required
for the field in upcoming weeks. The climatic variables are recorded from the
agro-meteorological station that are present near by the field. Temperature (T), Relative
humidity (RH), Global radiation (GR), Wind speed (WS), Rainfall (RF), Dew point
(DP), Vapour Pressure Deficit (VPD) [18] are some of the climatic data downloaded
from agro-meteorological stations from SIAM websites.

4 Conclusion

In this paper different machine learning techniques to predict soil moisture and the real
time use of IoT in agriculture is discussed. Though there are number of machine
learning algorithms, studies show that (SVM) provides reliable accuracy in predicting
soil moisture. In paper [3, 15, 16, 20] a comparative study between ML algorithms
were carried out, in which SVM showed an successful result. Training and testing
results shows that SVM model is able to better capture the interrelations among soil
moisture, backscatter, and vegetation than ANN and MLR models [20]. There are
many researches that has been carried in the field of agriculture using machine learning
algorithms. There is a need to further enhance IoT with machine learning techniques to
improve real time use of sensors in agriculture. IoT applications must be easy to use so
that the farmers can take the advantage of it. We need to develop applications to
analyse data captured by sensors in real time using Machine learning algorithms

References
1. Singh, D.A.A.G., Leavline, E.J., Priyanka, V., Swathi, V.: Agriculture classification system
using differential evolution algorithm. Int. Adv. Res. J. Sci. Eng. Technol. (IARJSET) 3, 24–
28 (2016)
2. Bhargavi, P., Jyothi, S.: Applying naive bayes data mining technique for classification of
agricultural land soils. Int. J. Comput. Sci. Netw. Secur. 9(8), 117–122 (2009)
3. Heung, B., Ho, H.C.D., Knudby, A., Bulmer, C.: An overview and comparison of machine
learning techniques for classification purposes in digital soil mapping. Geoderma 265, 62–77
(2016) ResearchGate
4. Krintz, C., Sethuramasamyraja, B., Liu, B.: SmartFarm: improving agriculture sustainability
using modern information technology. In: UCSB Tech Report (2016)
5. Rosa, D.D.L., Mayol, F., Moreno, F., Cabrera, F., Dıaz-Pereira, E., Antoine, J.: A
multilingual soil profile database (sdbm plus) as an essential part of land resources
information systems. Environ. Model. Softw. 17(8), 721–730 (2002). Elsevier Science
612 S.R. Juhi Reshma and A.S. Pillai

6. Rosa, D.D.L., Mayol, F., Moreno, F., Cabrera, F., Dıaz-Pereira, E., Antoine, J.: A land
evaluation decision support system (MicroLEIS DSS) for agricultural soil protection With
special reference to the Mediterranean region. Environ. Model. Softw. 19(10), 929–942
(2003). Elsevier Science
7. Demšar, D., Džeroski, S., Krogh, P.H., Larsen, T.: Using machine learning to predict the
impact of agricultural factors on communities of soil microarthropods. Metodološki zvezki
2, 147–159 (2005)
8. Lary, D., Alavi, A., Gandomi, A., Walker, A.: Machine learning in geosciences and remote
sensing. Geosci. Front. 7(1), 3–10 (2015). Elsevier Science
9. Elisabeth, N., Bui Brent, L.: Henderson., Viergever, K.: Knowledge discovery from models
of soil properties developed through data mining. Ecol. Model. 191(3), 431–446 (2005).
Elsevier Science
10. Hedley, C.P., Roudier, P., Yule, I.J., Ekanayake, J., Bradbury, S.: Soil water status and water
table depth modelling using electromagnetic surveys for precision irrigation scheduling.
Geoderma 199, 22–29 (2012)
11. Zhao, J., Zhang, J., Feng, Y., Guo, J.: The Study and Application of the IOT Technology in
Agriculture. IEEE (2010)
12. Ma, J., Zhou, X., Li, S., Li, Z.: Connecting Agriculture to the Internet of Things through
Sensor Networks. In: International Conferences on Internet of Things and Cyber Physical
and Social Computing IEEE (2011)
13. Nay, J., Burchfield, E., Gilliganc, J.: A Machine Learning Approach to Forecasting
Remotely Sensed Vegetation Health. arXiv (2016)
14. Kaur, K.: The idea of Smart villages based on Internet of Things (IoT). Int. Res. J. Eng.
Technol. (IRJET) 03(05), 1–4 (2016)
15. Lamorski, K., Pachepsky, Y., Sławiński, C., Walczak, R.T.: Using support vector machines
to develop pedotransfer functions for water retention of soils in poland. Soil Sci. Soc. Am.
(SSSAJ) 72, 1243–1247 (2007)
16. Shi, L., Duan, Q., Ma, X., Weng, M.: The research of support vector machine in agricultural
data classification. In: International Conference On Computer And Computing Technologies
In Agriculture, pp. 265–269. Springer (2011)
17. Murthy,A.S.R., Sudheer,Y., Mounika, K., Sravya Rao., Durga Prasad, P.: Cloud technology
on agriculture using sensors. Indian J. Sci. Techno. 9(17) (2016)
18. Navarro-Hellín, H., Martínez-del-Rincon, J., Domingo-Miguel, R., Soto-Valles, F.,
Torres-Sánchez, R.: A decision support system for managing irrigation in agriculture.
Comput. Electron. Agric. 124, 121–131 (2016)
19. Prashant, K.: Srivastava., Han, D., Ramirez, M.R. and Islam, T.: Machine Learning
techniques for downscaling SMOS satellite soil moisture using MODIS land surface
temperature for hydrological application. Water Resour. Manage. 27, 3127–3144 (2013)
20. Ahmad, S., Kalra, A., Stephen, H.: Estimating soil moisture using remote sensing data: a
machine learning approach. Adv. Water Resour. 33, 69–80 (2009)
21. Hu, S., Wang, H., She, C. and Wang, J.:AgOnt: Ontology for agriculture internet of things.
In: International Federation for Information Processing (IFIP), pp. 131–137 (2011)
22. Ghosh, S., Koley, S.: Machine learning for soil fertility and plant nutrient management using
back propagation neural networks. Int. J. Recent Innov. Trends Comput. Commun.
(IJRITCC) 2, 292 – 297. (2014)
23. Shruti, A., Jaishetty., Patil, R.: Iot sensor network based approach for agricultural field
monitoring & control Int. J. Res. Eng. Technol. 05 2016
 Impact of Machine Learning and Internet of Things in Agriculture 613

24. Ahmed, S., Shekhawat, A.S., Kumar, S.G. and Nair, M.K., Kumar, B. P.: ‘Intelligation’: an
IoT based framework for smarter irrigation. In: National Conference on Product Design
(NCPD). (2016)
25. Shi, Y., Wang, Z., Wang, X., Zhang, S.: Internet of things application to monitoring plant
disease and insect pests. In: International Conference on Applied Science and Engineering
(ASES 2015)