Performance evaluation of machinelearning techniques indiabetes prediction

Abstract

Diabetes diagnosis is very important at preliminary stage rather than treatment. In today’s world devices like sensors are used for detection of diabetes. Accurate classification techniques are required for automatic identification of diabetes disease. In regards to research diabetes prediction with minimal number of attributes (test parameters) is to be identified earlier research states about feature reduction but with less predictive accuracy. In this regards, this work exploits machine learning techniques(methodology) such as Logistic Regression (LR), Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF) and Neural Network (NN) with 10-fold Cross Validation (CV) for classification and prediction of diabetes with Feature Selection Methods (FSMs) using R platform. Above all models enable us to investigate the relationship between a categorical outcome and a set of explanatory variables. The experiment was conducted on PIMA Indian diabetes dataset selected from UCI machine learning repository. From the experimental results it is identified that for full set of diabetes dataset attributes, Classification Accuracy (CA) achieved was 84.25%whereas with reduced set attributes an accuracy of 85.24% is achieved using NN with 10-fold CV technique compared to others which will help in medical application to predict diabetes with minimal features.