Towards Personalized Healthcare: Explainable AI and Deep Learning for Lifestyle-Driven Health Classification

Abstract

Personalized healthcare, driven by lifestyle-informed prediction, has the potential to revolutionize preventive medicine by enabling early detection of health risks. This study proposes an integrated framework that combines machine learning (ML), deep learning (DL), and explainable AI (XAI) for lifestyle-driven health classification. The system processes lifestyle and health attributes including exercise frequency, body mass index (BMI), dietary habits, and sleep duration through a modular pipeline of preprocessing, predictive modeling, explainability, and evaluation. Experimental results on the curated dataset demonstrate that deep learning models significantly outperform traditional ML baselines. The LSTM achieved the best performance with an accuracy of 92.4%, F1-score of 0.91, and AUC of 0.95, while the DNN followed closely with 90.8% accuracy. In contrast, traditional models such as Logistic Regression and Random Forest achieved 82.3% and 85.6% accuracy, respectively. Beyond predictive performance, explainability analysis via SHAP identified exercise frequency, BMI, and sleep duration as the most influential global predictors, while LIME provided instance-level explanations, revealing how high physical activity mitigates risks associated with elevated BMI. These results highlight the dual advantage of robust predictive modeling and transparent interpretation. The proposed framework contributes to advancing trustworthy, personalized, and preventive healthcare, bridging the gap between accuracy and interpretability in clinical decision support.

Ähnliche Arbeiten