Enhanced LDA with SVM and Ensemble Learning for Advanced ADL Classification Using Smartwatch Data

Abstract

The demographic shift towards an older population introduces significant challenges for healthcare, necessitating advancements in predictive, dynamic care models. Central to this transformation are Activities of Daily Living (ADLs) and Instrumental Activities of Daily Living (IADLs), which are crucial for evaluating an individual's functional status. This study leverages Machine Learning (ML) with smartwatch technology for continuous, non-intrusive monitoring of ADLs /IADLs. By integrating Support Vector Machine (SVM) and Ensemble Learning (EL) with enhanced Linear Discriminant Analysis (LDA), we aim to accurately classify a broad spectrum of daily activities. Data from 30 participants, captured through a smartwatch, underwent analysis to assess these methodologies' efficacy in real-life applications. Results indicate that LDA enhanced classification accuracy, with EL models demonstrating superior performance in recognizing complex activities—achieving over 90% accuracy. The study also examines the hardware performance of these models, noting variations in RAM usage, inference times, CPU usage, and network sizes. EL models, especially when coupled with LDA, offer an optimal balance between computational demands and classification precision. This investigation underlines the potential of integrating sophisticated data processing techniques with wearable technologies for health monitoring, presenting a viable approach to overcoming real-life deployment challenges and advancing personalized healthcare for the aging demographic.