Early Detection of Alzheimer’s Disease from Daily Behavioral Time-Series Using Interpretable Hidden Markov Models

Abstract

Alzheimer's disease is a progressive neurodegenerative disorder that demands early, accessible, and interpretable diagnostic approaches. Traditional diagnostic methods, such as neuroimaging and clinical testing, are often costly and impractical for continuous monitoring. In this paper, we propose a novel diagnostic system that employs Hidden Markov Models (HMMs) to analyze simulated daily behavioral time-series data inspired by smart-home sensor patterns. The model captures transitions between cognitive states (Normal, MCI, and Early Alzheimer's) based on observed Activities of Daily Living (ADLs). A synthetic dataset (Sim-ADL), generated to reflect real-world behavioral distributions like the CASAS-Aruba environment, is used to train and validate the system. The framework demonstrates moderate classification performance," revealing potential challenges in identifying early cognitive decline stages. Despite this, the model provides clinically interpretable results via transition and emission probability matrices. Compared to deep learning methods, the HMM approach balances interpretability and performance, making it a suitable low-cost tool for early-stage Alzheimer's detection and continuous home-based cognitive monitoring.