Book Description

The average age of people has increased due to advances in health sciences, which has led to an increase in the elderly population. This is positive news, but it also raises questions about the quality of independent living for older people. Clinicians use Activities of Daily Living (ADLs) to assess older people's ability to live independently. In recent years, portable computing devices have become more present in our daily lives. Therefore, a software system that can detect ADLs based on sensor data collected from wearable devices is beneficial for detecting health problems and supporting health care. In this context, this book presents several machine learning-based approaches for human activity recognition (HAR) using time-series data collected by wearable sensors in the home environment. In the first part of the book, machine learning-based approaches for atomic activity recognition are presented, which are relatively simple and short-term activities. In the second part, the algorithms for detecting long-term and complex ADLs are presented. In this part, a two-stage recognition framework is also presented, as well as an online recognition system for continuous monitoring of HAR. In the third and final part, a novel approach is proposed that not only solves the problem of data scarcity but also improves the performance of HAR by implementing multitask learning-based methods. The proposed approach simultaneously trains the models of short- and long-term activities, regardless of their temporal scale. The results show that the proposed approach improves classification performance compared to single-task learning.