Enhancing Diabetes Prediction Based on Pair-Wise Ensemble Learning Model Selection

Diabetes, a multifaceted health condition, presents significant challenges in terms of early diagnosis. This has sparked an interest in investigating the efficacy of two-combination ensemble learning (EL) techniques, which have shown potential in heightening the precision of predictions. Although methodologies such as logistic regression (LR), K-nearest neighbors (KNN), naive bayes (NB), support vector machine (SVM), decision tree (DT), and random forest (RF) have been promising when applied in isolation, their collective strength through an ensemble approach is not comprehensively studied. This investigation fills this research void by evaluating different integrations of these methodologies to construct solid predictive systems for diabetes using the pima indian diabetes dataset (PIDD). This approach employs ensemble learning, a strategy that amalgamates several analytical models to improve the reliability of predictions. By strategically synthesizing models like LR, KNN, NB, SVM, DT, and RF, this study aims to utilize the unique advantages that each model offers. Among these combinations, the SVM+RF, KNN+DT, and SVM+DT configurations stand out, delivering impressive accuracy rates of 84 %, 84 %, and 83 % respectively on tests. The effectiveness of the SVM+DT combination is further highlighted through receiver operating characteristic (ROC) curve analysis, showing extraordinary discrimination power with the highest area under the curve (AUC) score reaching 0.9. When comparing these results to findings from contemporary research, the innovative combinations proposed here-namely, SVM+DT, KNN+DT, and SVM+RF-provide enhanced performance, surpassing recent models by margins of 6 %, 6 %, and 4.4 % accordingly. These results underline the advantage of integrating a variety of classifiers to amplify the accuracy of diabetes predictions.