Supervised and Unsupervised Learning on Human Activity Sequences: A Novel Study of Influential Dimensions on Physical Activity Trajectories among Adolescents in Qatar

Many studies have been conducted on Human Activity Recognition (HAR) which main objective is to recognize the movements of human from a series of observations on the people's activities and the surrounding environments. Other researches focus on Human Activity Monitoring (HAM) which aims to measure the changes in physiological parameters of people to detect the normal from abnormal behaviors. Both, Recognition and Monitoring, work with the original sensory data (time series). However, very little is currently known about Human Activity Analysis (HAA) which works on the recognized sets of activities (sequences of activities). These types of studies can help in answering many pertinent research questions related to public health and wellbeing such as sleep quality, obesity problems, and diabetes. For example, much uncertainty still exists about the relationship between human activity and quality of sleep among all ages. This thesis is the first study to undertake a longitudinal multi-dimensions analysis of human physical activities sequences (trajectories). It is mainly designed to explore and analyze the association between human physical activity and quality of sleep. Additionally, after thorough analysis, the relational study extended to include other variables such as age, gender, and BMI interpretation. This thesis introduces an efficient methodology for analyzing and classifying human physical activities using machine learning. This methodology consists of a framework, for identifying the looked-for dimension in the physical activities, and complemented by a four steps procedure which help in measurement selection and conducting the analysis. Accordingly, some topologies can be discovered among sets of activities (sequences of activities) which can be then correlated with different variables such as the quality of sleep.