Employing Random Forests and Long Short-Term Memory Networks for Enhanced Predictive Modeling of Disease Progression

Abstract

This research paper explores an innovative approach to predictive modeling of disease progression by integrating Random Forests (RF) and Long Short-Term Memory (LSTM) networks. The study leverages the strengths of RF in handling structured tabular data and LSTM's capability in processing sequential data, aiming to enhance the accuracy and reliability of disease progression forecasts. We employ a hybrid model that synergistically combines these techniques to capture intricate patterns in large and complex healthcare datasets. The research utilizes publicly available datasets on chronic diseases, focusing on conditions with significant sequential data, such as diabetes and cardiovascular diseases. The model's performance is evaluated against traditional methods, demonstrating superior predictive accuracy and robustness across various metrics, including RMSE, MAE, and ROC-AUC. The integration strategy involves training an RF model to identify important features and an LSTM network to model temporal dependencies, subsequently combining their outputs for final prediction. Our results reveal that the hybrid model effectively handles missing data and variable-length inputs, offering scalable solutions for real-world applications. This study underscores the potential of combining ensemble learning with deep learning architectures to advance predictive analytics in healthcare, providing a framework that could inform clinical decision-making and personalized treatment plans. Further research will focus on optimizing computational efficiency and exploring the generalizability of this approach across diverse medical conditions.