Human Behaviour Recognition of Elderly in Single-Resident IoT Enabled Smart Homes: An Applied Machine Learning Approach

Sammanfattning: In Human Activity Recognition (HAR) systems, activities of daily living/human be-haviour are recognized using sensor data by applying data mining techniques and machinelearning algorithms to the collected data. This allows for customised and automated ser-vices to support humans’ daily living. Along with recognizing human behaviour, HARcan provide insights into the abnormal behaviour of individuals that might link to possi-ble health conditions. Health monitoring applications are widely used for patients withchronic diseases, especially to give feedback to the users and promote self-awareness, forexample, persons with dementia who need constant monitoring to minimize the risk ofundesirable events. In developing HAR applications and services along with anomalydetection, obstacles and challenges remain that need to be addressed and motivate ourresearch focusing on older adults living in real-world conditions. The need for care hasincreased, especially with the demographic developments and growing population of 65years and over. Some of the challenges identified are installing IoT devices and sensors,collecting data and maintaining the system in an uncontrolled environment, and iden-tifying valuable features to extract from sensors while ensuring the development of anon-intrusive and privacy-preserving system. Our research uses smart home sensors toinfer ADLs (e.g., eating, sleeping, and bathroom visits) for older adults in single-residenthomes. The research aims to develop and validate an anomaly detection based IoTsystem within HAR to support older adults in living longer independently and for therelatives (caregivers) to provide the necessary support. In addition, the system will helpaddress the challenges of the ageing population and the increased burden on healthcareresources. This thesis identifies technologies and datasets suitable for IoT-enabled smarthealthcare applications to recognize near real-time and short- and long-term behaviouralchanges.We propose to develop a life conditions model for each individual by understandingthe routines and the activities of daily living based on interviews with older adults andtheir caregivers and the usage of collected datasets with a focus on motion sensors, todevelop a life condition model for each individual to recognize human behaviour basedon context information such as time and location to analyze overall behavioural change.Hence, building an anomaly detection system that preserves privacy for near real-timebehavioural changes and supports large-scale deployment. Our research methodologyfollows a quantitative research methodology and is slightly modified to include qualitativebased on interviews.We defined activity models based on contextual information such as time and locationto extract features suitable for inferring daily living activities,  model behavioural pat-terns, and, after that, detect abnormal activities in each daily routine by utilizing motionsensors as suitable sensing devices for non-intrusive IoT-enabled smart healthcare appli-cations in single-resident homes. For example, we applied a statistical method to build aroutine or habit model for each older adult. We utilized unsupervised clustering methodsK-means and LOF and reinforcement learning sleeping to recognise sleep activity pat-terns and detect anomalies. Further, to recognise all variations of the person’s behaviourand detect short and long-term behavioural changes in older people’s daily behaviour,we applied LSTM and VAE algorithms. In addition, we developed an anomaly detectionsystem that preserves privacy to support large-scale deployments by utilizing federatedlearning to build a generalizable model that learns from different experiences and appliesthe knowledge from other persons.vi