Enhancing Diabetes Prediction with a Random Forest Ensemble of Extreme Learning Machine and Long Short-Term Memory Models

This study introduces a novel hybrid ensemble approach for enhanced diabetes prediction, combining the strengths of an Extreme Learning Machine (ELM), a Long Short-Term Memory (LSTM) network, and a Random Forest (RF) classifier. The proposed framework addresses limitations of existing models by leveraging ELM’s rapid training, LSTM’s capacity to capture temporal dependencies (although less critical for this dataset), and RF’s robust predictive power. Specifically, ELM and LSTM models are trained independently, and their predictions are then used as input features for the RF classifier. Evaluated on the widely used Pima Indians Diabetes Dataset, the ensemble model achieves a remarkable 93% accuracy and an Area Under the Curve (AUC) of 89%, significantly outperforming individual models and demonstrating improvements over previous hybrid approaches. This optimized balance between computational efficiency and predictive accuracy makes the proposed ensemble highly suitable for practical clinical applications in diabetes diagnosis and management.