The objective of this research is to develop a cyber-physical system composed of accelerometers and novel machine learning algorithms to analyze data in the context of a set of driving health care applications. The approach is to develop novel machine learning algorithms for temporal segmentation, classification, and detection of subtle elements of human motion. These techniques will allow quantification of human motion and improved full-time monitoring and assessment of medical conditions using a lightweight wearable system. The scientific contribution of this research is in advancing machine learning and human sensing in support of improved medical diagnoses and treatment monitoring by (i) modeling human activity and symptoms through sensor data analysis, (ii) integrating and fusing information from several accelerometers to monitor in real-time, (iii) validating the efficacy of the automated detection through assessments applying the state of the art in diagnostic evaluation, (iv) developing novel machine learning methods for temporal segmentation, classification, and discovery of multiple temporal patterns that discriminate between temporal signals, and (v) providing quality measures to characterize subtle human motion. These algorithms will advance machine learning in the area of unsupervised and semisupervised learning. The driving applications for this research are job coaching for people with cognitive disabilities, tele-rehabilitation for knee osteo-arthritis, assessing variability in balance and gait as an indicator of health of older adults, and measures for assessing Parkinson's patients. This research is highly interdisciplinary and will train graduate students for careers in developing technological innovations in health and monitoring systems.