A Potential Research for Maximizing Predictive Power in Neural Networks through the Integration of Feature Engineering in Diabetes Prediction

In the realm of critical diseases, diabetes stands prominent, afflicting numerous individuals with its multifaceted origins. Various factors contribute to the onset of diabetes, amplifying the risk of associated conditions like heart conditions, eye issues, and neurological impairment. Within this healthcare landscape, the imperative role of predictive modelling unfolds. Machine learning, harnessed in healthcare industries with voluminous databases, becomes a potent tool. This research study explores an innovative framework, the Machine Learning-based Diabetic Prediction Model (MLDPM), aiming to enhance the predictive capabilities of neural networks in diabetes prediction. It explores diverse neural network architectures and feature selection techniques, optimizing their collaboration for high accuracy and interpretability.The ML-DPM meticulously explores the optimal combination of neural network architecture and feature selection methods to maximize predictive power in diabetes prediction using the UCI data repository. Through an extensive experimental study, specific neural networks demonstrate superior performance with particular feature selection techniques. The Narrow Neural Network achieves a high accuracy of 98.1% with an integration of ANOVA and Kruskal, while the Wide Neural Network performs better with MRMR 99.4%. Bilayered Networks exhibit improved performance (96.2%) with integration of Kruskal Wallis. Notably, the Optimizable Neural Network, coupled with ANOVA, demonstrates superior accuracy at 100%. These findings provide valuable insights into diabetic prediction, advancing predictive modeling in healthcare.