A Unified Deep Learning Framework for Predictive Healthcare Analytics Using TJO-Based Feature Selection and Tree Growth Optimized Graph-Temporal Modeling

Abstract

The intersection of healthcare and biomedical informatics has witnessed exponential data growth through electronic health records (EHRs), wearable devices, and multi-modal biomedical sensors. However, extracting actionable intelligence from this complex, high-dimensional, and temporally evolving data remains a major challenge. This paper proposes a unified deep learning framework that integrates Tree Growth Optimization (TGO)-based feature selection with a Tree Growth Optimized Long Short-Term Memory (TGO-LSTM) and graph-based modeling for predictive healthcare analytics. The framework efficiently handles temporal dependencies and inter-patient relationships by combining graph and sequential learning paradigms. TGO plays a dual role—selecting the most informative biomedical features and tuning LSTM hyperparameters for optimal predictive accuracy. The proposed model was validated using benchmark healthcare datasets such as MIMIC-III and PhysioNet, demonstrating superior disease prediction accuracy, interpretability, and computational efficiency compared to existing CNN-LSTM and GNN-RNN models. Experimental results confirm that integrating Metaheuristic optimization with graph-temporal architectures enhances the predictive power of healthcare systems while maintaining scalability and transparency—key requirements for real-world clinical deployment.