Classifying Human Behaviors, Activities and Contexts from Mobile Sensor Data
This dissertation concerns applications of machine learning to time series classification. In particular, it investigates methods for classifying human behaviors, activities, and contexts from mobile sensor data. First, it presents a system for classifying five behaviors (sitting, standing, walking/running, riding in a vehicle, and bicycling) from body-worn accelerometers and GPS devices. The advantages and disadvantages of placing the accelerometer on the subjects wrist or hip are discussed. The system is based on a two-level classifier consisting of a random forest and hidden Markov model. Second, it presents a system for automatic classification of a variety of rich context labels using the sensors built into mobile phones and smartwatches. This system uses multilabel classification and investigates several methods of sensor fusion.