Identification of Influential Features for Diabetes Prediction at Early Stage

Early prediction of diabetes can be a vital discovery in the field of diabetic treatment. Patients can be alarmed at an early stage and take proper initiatives to evade critical health situations. The clinical process of diabetes prediction at an early stage is costly and time-worthy. But, a proper selection of clinical features can elucidate the problem simply. In the near past, many machine learning(ML) researchers have studied this field. Unfortunately, most of them were concerned with the improvement of classification. We have concentrated on the feature section for diabetes prediction and the enhancement of classification in this study. Our proposed technique has found the top 5 features that can result in a practically perfect classification rate of 99.51%. In the data pre-processing stage, we have used Min-Max normalization for data normalization and iForest for outlier removal. The Random Forest based Recursive Feature Elimination (RFE-RF) test, the Minimum Redundancy Maximum Relevance (mRMR) test, and the Chi-Square test have been utilized in this research as the feature selection methods. Also, the Support Vector Machine (SVM), the K Nearest Neighbors (KNN), and the Naive Bayes (NB) have been utilized as the classifiers in this research.