UCLA

Recent advances in electronic miniaturization, sensor technology, and wireless communications have opened the possibility of ubiquitous, small, and low-power sensor nodes that gather, process, and transmit various data over a long period of time. One of the key applications for these new technologies is in the area of remote health monitoring. Having the ability to continuously monitor numerous bodily measurements, as opposed to the occasional on-the-spot examination performed at a doctor visit, can potentially revolutionize the health care system. For the first time, physicians can make more informed decisions based the continuous history of a patient's wellness, rather than relying on the incomplete snapshots from the traditional medical records.

More importantly, using advanced data mining and machine learning techniques, it is possible to discover a wealth of patterns, knowledge, and relationships based on the data collected from a large population of different background, ethnicity, age group, and medical history. This thesis focuses specifically on event predictions for remote health monitoring. These events can be either acute clinical episodes, such as falling and epileptic seizure, or chronic conditions, such as congestive heart failure and diabetes. Based on the different requirements, this thesis tackles four key issues of event predictions for remote health monitoring: (a) Subsequence-based prediction, (b) Sequential pattern mining, (c) Precursor pattern discovery, and (d) Predictions using discrete data. For each issue, one or multiple application areas have been identified, and the proposed algorithms have been validated using data gathered from real patients.