Association rule mining based disease signatures discovery

The focus of this thesis is discovering disease signatures using association rule mining. A disease signature is the common characteristics that may give the disease patients a distinct identity. In this thesis, we are specifically analyzing the dynamic gaits of Parkinson's patients and healthy controls to identify potential disease signatures. This can be useful as such signatures can be considered potential indicators of the disease. Since there is no preemptive test or marker for Parkinson's disease diagnoses, these types of signatures can help physicians be proactive. Currently a diagnosis of Parkinson's disease is primarily based on a patient's medical history and some neurological exams which include analysis of movement and balance. Our approach can be generalized to any type of gait analysis.;We use association rule mining as a building block to identify signatures. We tested our approach on the dynamic gaits of 15 Parkinson patients and 16 healthy controls. We used Coron and Weka systems in our implementation. The data was obtained from PhysioBank, a public source for biomedical data.;Our results identified significant signatures for Parkinson's disease in approximately 50% of the data. We also identified some overlap in signatures between Parkinson's patients and healthy controls, indicating that gait speed and height may be important factors in discovering signatures. We validated our results using Pearson correlation and found that correlation increases substantially in Parkinson's patients, indicating that the signatures are valid, especially in Parkinson's patients.