An Optimized Hybrid Ensemble Approach forAccurate Diabetes Prediction Using SVM, RF,and Neural Networks

Abstract

Diabetes constitutes a significant worldwide health concern,underscoring the necessity for early identification and intervention. There are encouraging opportunities to create trustworthy models for diabetes classification using machine learning approaches. Medical professionals can use machine learning to find hidden aspects and patterns in large healthcare datasets, enabling them to make well-informed decisions about patient projections. Current diabetes classification algorithms have accuracy problems despite advancements. We present a novel hybrid machine learning approach in this work that blends support vector machines, random forests, and neural network classifiers. Our model is trained and tested on a large diabetic patient dataset using the holdout and k-fold cross-validation methods. On the holdout set, 98.43% accuracy was achieved by the hybrid method, while k-fold cross-validation yielded 96.4%. The hybrid model outperformed standalone classifiers: SVM (84.45%), Random Forest (94.50%), and Neural Network (96.12%). The algorithm's ability to predict diabetes is demonstrated by tests based on measures of recall (0.98), precision (0.98), and F1 score (0.98), which make it a useful tool for early identification and intervention.