DIABETES DATA HANDLING BY ML ALGORITHMS: A COMPARATIVE STUDY

The functioning of five standard ML approaches for predicting diabetes is evaluated in this study utilizing a publicly accessible data set through Kaggle that contains pertinent medical and socioeconomic variables. To solve values that were not present, outliers, and a notable inequalities in classes (32:1 proportion of people without diabetes to mellitus samples), which presents serious obstacles to model performance; extensive preprocessing operations were carried out. Important metrics for performance such as optimal balancing recall, precision,  accuracy, a standardized F1 scores, specificity, sensitivity, and rate of detection were used to evaluate the algorithmic methods—  Artificial Neural Network (ANN), Naive Bayes (NB), Support Vector Machine (SVM), Random Forest (RF),  Decision Tree (DT) and Logistic Regression (LR). RF outperformed the other models in terms of performance in prediction, achieving the greatest optimal balance (93%), then a DT (82%) and SVM (83%). The results highlight how algorithms for learning can improve diabetic detection and treatment, and they suggest that algorithm choice must be guided by an appropriate balance of effectiveness indicators, particularly when working with unbalanced medical data.

Keywords: Diabetes, ML, Evaluation of Performance, SVM, ANN, DT, LR, Naïve Bayes, Random Forest   

Diabetes ML Evaluation of Performance SVM ANN DT LR Naïve Bayes Random Forest