Performance Evaluation of Various Classifiers for Diabetes Detection: A Comparative Approach

Abstract

Various Supervised learning algorithms or techniques viz. Random Forest, Naïve Bayes Classifier, LogisticRegression (LR), K-Nearest Neighbour (KNN) algorithm, Support Vector Machines (SVM), etc are used for thepurpose of data classification.. But the question is which of the classification technique accurately identifies thissensitive disorders like Diabetes. The accuracy, specificity and sensitivity, are some of the important performanceevaluation parameters, which are required to be analysed for every machine learning algorithm. In the performedwork, the various classification techniques viz. (NB) Naïve Bayes Classifier, (LR) Logistic Regression, (KNN)K-Nearest Neighbour algorithm, (SVM) Support Vector Machines, and (RF) Random Forest are compared onthe basis of the accuracy, sensitivity and specificity as the performance evaluation parameters. The classifierswere exposed to the Pima Indian dataset for classification of diabetes, and their respective performance metricsAccuracy, Sensitivity, and Specificity were compared. It is found that on account of accuracy sensitivity andspecificity the Random Forest performed the best on the Pima Indian Dataset for the Diabetes detection.