Metaheuristic Optimization-Assisted LSTM-DBN Model for Predictive Healthcare Analytics and Biomedical Informatics

Abstract

The integration of artificial intelligence (AI) with healthcare informatics has led to significant advancements in disease prediction, patient risk assessment, and medical decision support. However, healthcare data is highly heterogeneous, dynamic, and prone to imbalance and uncertainty—posing challenges for accurate temporal modeling and feature extraction. This research presents a Metaheuristic Optimization-Assisted LSTM-DBN Model, a hybrid deep learning framework designed to enhance predictive healthcare analytics by fusing Long Short-Term Memory (LSTM) networks for temporal modeling with Deep Belief Networks (DBN) for hierarchical feature abstraction. To optimize performance, Tree Growth Optimization (TGO) and Tunicate Swarm Optimization (TSO) algorithms are employed for dynamic hyperparameter tuning. The model is tested on benchmark biomedical datasets, including physiological signal sequences, patient vital trends, and disease progression records. Experimental outcomes demonstrate superior prediction accuracy, stability, and interpretability compared to traditional deep models. This optimized hybrid architecture enables precise temporal analysis of patient health trajectories, facilitating early diagnosis, clinical decision support, and efficient biomedical data modeling—paving the way for intelligent, data-driven healthcare systems.