The document discusses using machine learning algorithms like Naive Bayes, random forest and decision tree to predict anemia disease using complete blood count data. It finds that Naive Bayes technique performs best in terms of accuracy compared to decision tree and random forest.
The document discusses using machine learning algorithms like Naive Bayes, random forest and decision tree to predict anemia disease using complete blood count data. It finds that Naive Bayes technique performs best in terms of accuracy compared to decision tree and random forest.
The document discusses using machine learning algorithms like Naive Bayes, random forest and decision tree to predict anemia disease using complete blood count data. It finds that Naive Bayes technique performs best in terms of accuracy compared to decision tree and random forest.
The document discusses using machine learning algorithms like Naive Bayes, random forest and decision tree to predict anemia disease using complete blood count data. It finds that Naive Bayes technique performs best in terms of accuracy compared to decision tree and random forest.
See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/329484705
Machine Learning Algorithms for Anemia Disease Prediction: Select
Proceedings of IC3E 2018
Chapter · January 2019
DOI: 10.1007/978-981-13-2685-1_44
CITATIONS READS
17 4,520
3 authors, including:
Manish Jaiswal T.J. Siddiqui
University of Allahabad University of Allahabad 6 PUBLICATIONS 18 CITATIONS 53 PUBLICATIONS 571 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Speech, Image, and Language Processing for Human Computer Interaction: Multi-Modal Advancements View project
recommender system View project
All content following this page was uploaded by Manish Jaiswal on 01 April 2019.
The user has requested enhancement of the downloaded file.
Machine Learning Algorithms for Anemia disease Prediction
Manish Jaiswal1, Anima Srivastava2, and Tanveer J. Siddiqui3
1 Department of Electronics & Communication, University of Allahabad, manish.jk50@gmail.com 2,3 Department of Electronics & Communication, University of Allahabad {animasparklestar,siddiqui.tanveer}@gmail.com
Abstract. The remarkable advances in health industry have led to a significant
production of data in everyday life. This data requires processing to extract use- ful information, which can be useful for analysis, prediction, recommendations, and decision-making. Data mining and machine learning techniques are used to transform the available data into valuable information. In medical science, dis- ease prediction at the right time is the central problem for professionals for pre- vention and effective treatment plan. Sometimes, in absence of accuracy this may lead to death. In this study, we investigate supervised machine learning algorithms - Naive Bayes, random forest, and decision tree algorithm - for pre- diction of anemia using CBC (complete blood count) data collected from pa- thology centers. The results show that Naive-Bayes technique outperforms in terms of accuracy as compared to C4.5 and random forest.
The modern health care system generates huge volumes of data every day. There is a need to mine and analyze these data to extract useful information and to reveal hid- den pattern. Data mining is the process of discovering new patterns from data collect- ed from varying sources. A number of machine learning algorithms have been used successfully in making prediction in various domains such as healthcare, weather forecasting, stock price prediction, product recommendation. An important aspect of medical science research is the prediction of various diseases and factors that cause them. In medical domain, healthcare data are being used to predict epidemics, to de- tect disease, to improve quality of life and avoid early deaths [1]. In this work, we investigate three different classification algorithms for its prediction. Anemia is defined as decrease in amount of red blood cells (RBCs) or hemoglobin in the blood [2] that has significant adverse health consequences, as well as adverse impacts on economic and social development. Although the most reliable indicator of anemia is blood hemoglobin concentration, but there are a num- ber of factors that can cause anemia such as iron deficiency, chronic infections such as HIV, malaria, and tuberculosis, vitamin deficiencies, e.g. vitamins B12 and A, cancer, and acquired disorders that affect red blood cell production and hemoglobin synthesis. Anemia causes fatigue and low productivity [3, 4, 5] and, when it occurs in preg- nancy, may be associated with increased risk of maternal and perinatal mortality [6, 7]. According to World health organization (WHO), maternal and neonatal mortality were responsible for 3.0 million deaths in 2013 in developing countries. Anemia disease prediction plays a most important role in order to detect other as- sociated diseases. Anemia disease is classified on the basis of morphology or on the basis of its underlying cause (Figure 1). Based on the morphology, anemia is divided into three types, which are normo- cytic, microcytic and macrocytic. Based on cause, anemia is classifies into three types namely blood loss, inadequate production of normal blood and excessive de- struction of blood cells.
Fig. 1. Classification of Anemia
In this paper, we attempt to investigate the performance of Naive-Bayes, random
forest, and decision tree algorithm for anemia disease prediction on dataset collected from local pathology centers. The need of this investigation arises from the fact that the underlying cause of the disease varies from one region to another. Although, ran- dom forest classifier has been earlier investigated for predicting heart and chronic kidney disease but to the best of our knowledge it has not been investigated for ane- mia disease prediction. This adds novelty to the work.
The rest of the paper is organized as follows:
Section 2 introduces briefly reviews existing related work. In section 3, we discuss various types of anemia diagnosis tests. Section 4 presents proposed methodology. Section 5 presents experimental details and discussion. Finally, we conclude in sec- tion 6. 2 Related Works
In the last decade, numerous data mining and machine learning techniques have been used for anemia disease. Most noted once are the following: In [8], SMO support vector machine and C4.5 decision tree algorithm has been used for the prediction of anemia and a performance comparison of the two algo- rithms is done. In [11], WEKA is used to get a suitable classifier for developing a mobile App, which can predict and diagnose Hematological data comments. The authors com- pared neural network classification algorithms with J48 and Naïve Bayes classifier. The results show that J48 classifier exhibits maximum accuracy. Dogan & Turkoglu [12] developed a decision support system for detecting Iron- Deficiency Anemia using decision tree algorithm. The algorithm uses three hematol- ogy parameters, Serum iron, Serum iron-binding capacity and Ferritin. The evaluation is done on Data of 96 patients and the results were successfully matched with Physi- cian’s decision. Abdullah and Al-asmari[13] experimented with WEKA algorithms: Naive-Bayes, Multilayer Perception, J48 and SMO in an attempt to predict anemia types using CBC reports. The evaluation was done on real data constructed from CBC reports of 41 anemic persons. Similar to [11], J48 decision tree algorithm along with SMO was the best performer with an accuracy of 93.75%. Unlike the work in [11] and [13], we have chosen a different set of classifier and local data in our work.
3 Diagnostic tests Classification
There are four main tests that are ordered to diagnose anemia disorder which are complete blood count (CBC), ferritin, PCR (Polymerase chain reaction ) and hemo- globin electrophoresis.
CBC test is the most frequently blood test to measure overall health and determine a wide range of diseases [8] including anemia, infection and leukemia. A com- plete blood count test measures almost 15 tests including: hemoglobin (Hb), Red blood cells (RBC), hematocrit (HCT), mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV), and so on [8]. A ferritin test measures the amount of iron store in the body. High levels of ferritin indicate an iron storage disorder, such as hemochromatosis. Low levels of ferritin indicate iron deficiency, which causes anemia. PCR test is a molecular test, which is used to diagnose genetic disorder. A hemoglobin electrophoresis test is a blood test used to measure and identify the different types of hemoglobin in the bloodstream. 4 Methodology
We have used three classifiers namely Random forest, Naive – Bayes and Decision tree C4.5 algorithm. Figure 2 depicts the flowchart of the proposed method.
4.1 Random Forest Algorithm
Random forest (RF) algorithm derives from decision tree classifier. It is a combina- tion of tree predictors which aggregates the results of all the trees in the collection and uses majority voting in prediction.
4.2 Decision Tree Algorithm
A decision tree is a tree in which each branch node represents a choice between a number of alternatives, and each leaf node represents a decision. It has been exten- sively used in various fields [9] [10]. C4.5 (J48 in WEKA) is a Decision Tree devel- oped by Ross Quinlan.
4.3 Naive – Bayes Algorithm
Naive – Bayes Algorithm is based on Bayes rule of conditional probability. It uses all the attributes contained in the data and analyzes them individually as though they are equally important and independent of each other. It requires very less amount of train- ing data.
Data Collection
Pre-processing
Classifier Learning
Anemia Disease Prediction
Performance Evaluation
Fig. 2. Flowchart of Proposed Model
5 Experimental Results and Discussion
5.1 Dataset We collect data from different pathology centre and laboratory test centers in nearby area. The collected dataset consists of 200 test samples. These are CBC test data. The dataset contains 18 attributes out of which we have selected only those, which are required for anemia disease detection. These are Age, Gender, MCV, HCT, HGB, MCHC and RDW
5.2 Experimental setup
The proposed method uses CBC test values. First, the data is pre-processed to extract the seven attributes as mentioned in 5.1. Then, we apply the random forest, decision- tree and NB classifier on it. The performance evaluation is done in terms of accuracy and mean absolute error (MAE). The mean absolute error (MAE) is measures how close the predictions are to the eventual outcomes. Table 1 shows the results of the three classifiers. Ten Fold cross validation has been used to obtain accuracy.
The comparative performance of each classification algorithm based on accuracy and
MAE is shown in Figure 3 and Figure 4. The Naive-Bayes classifier exhibits best performance on our dataset, which is unlike [11] and [13]. It is not surprising because the dataset being used in these works are different and the cause of disease in different countries might be different. We achieve a maximum accuracy of 96.09% with NB classifier which is better than the best performing classifiers- SMO and J48 with an accuracy of 93.75% - reported in [13]. Mean Abosute Error 0.035 0.0345 0.034 0.0335 0.033 (MAE) 0.0325 0.032 Random C4.5 Naïve- Bayes Forest
Fig. 4. Comparison of Accuracy using each Algorithm
6. Conclusion & Future Work
In this paper, we have compared the performance of three different classifiers in the prediction of anemia disease. The experimental result on a sample dataset suggests that Naive- Bayes classification algorithm provides best performance in terms of ac- curacy as compared to C4.5 and Random forest. Automatic prediction can reduce manual effort involved in diagnosis. In future, automated tools can be developed which can helps the prediction results to suggest further diagnosis. Such automated tools can prove valuable in timely detection of more serious disease. Furthermore, such disease prediction system can extended to recommend a treatment plan.
Reference 1. Arun, V, et al.: Privacy of Health Information in Telemedicine on Private Cloud, Interna- tional Journal of Family Medicine & Medical Science Research. (2015) 2. Provenzano, R., Lerma, E.V., & Szczech, L.: Management of Anemia. Springer.(2018) 3. Ezzati, M., Lopez, Ad., Rodgers, A., Murray, C.J.L.: Comparative quantification of health risks: global and regional burden of disease attributable to selected major risk factors. Ge- neva: World Health Organization. (2004) 4. Balarajan, Y., et al.: Anaemia in low-income and middle-income countries. (2011) 5. Haas, J.D., Brownlie, T.: Iron deficiency and reduced work capacity: A critical review of the research to determine a causal relationship. J Nutr. (2001) 6. Kozuki, N., Lee, A.C., Katz, J.: Child Health Epidemiology Reference Group. Moderate to severe, but not mild, maternal anemia is associated with increased risk of small-for- gestational-age outcomes. J Nutr. (2012) 7. Steer, P.J.: Maternal hemoglobin concentration and birth weight. Clin Nutr. (2000) 8. Shilpa A. Sanap, Meghana Nagori, Vivek Kshirsaga.: Classification of Anemia Using Data Mining Techniques.: Swarm, Evolutionary, and Memetic Computing pp 113-121. Springer (2011). 9. Jerez-Aragonés J.M. et al.: A combined neural network and decision trees model for prog- nosis of breast cancer relapse. Artif Intell Med. (2003) pp 45–63. 10. Podgorelec, V. et al.: Decision trees: an overview and their use in medicine. J Med Syst. (2002) pp: 445–463 11. N. Amin and A. Habib Comparison of different classification techniques using WEKA for hematological data, American Journal of Engineering Research, Volume-4, Issue-3, pp-55- 61 (2015) 12. Dogan, S., Turkoglu, I.: Iron deficiency anemia detection from hematology parameters by using decision tree. International journal of Science and technology. (2008) pp: 85-92. 13. Manal Abdullah and Salma Al-Asmari, Anemia types prediction based on data mining classification algorithms, Communication, Management and Information Technology, (2016) Taylor & Francis Group, London,
