Optimized Machine Learning Models For Early Detection Of Alcohol Use Disorder: A Hybrid Approach

Abstract

Alcohol Use Disorder (AUD) remains a critical global health issue, necessitating the development of advanced diagnostic systems for early and accurate detection. Conventional diagnostic methods often exhibit subjectivity and limited predictive capability, emphasizing the need for intelligent computational techniques. This research introduces a hybrid optimization framework that integrates machine learning models with metaheuristic optimization strategies to enhance AUD detection. By incorporating evolutionary algorithms, such as Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), alongside deep learning techniques, the proposed approach optimally selects features, fine-tunes hyperparameters, and reduces overfitting. A diverse dataset combining clinical, behavioral, and neuroimaging data is used to train and validate the model, ensuring broad applicability across different populations. Comparative evaluations with traditional machine learning models indicate that the hybrid-optimized method substantially improves classification accuracy, sensitivity, and specificity in differentiating AUD from non-AUD cases. Additionally, explainable AI techniques are utilized to improve model interpretability, aiding healthcare professionals in understanding key predictive factors. The results highlight the potential of hybrid optimization in machine learning for AUD diagnosis, contributing to more reliable, data-driven clinical decision-making and early intervention strategies.