Exploratory and directed analysis of medical information via dynamic classification trees

摘要

Medical data are often voluminous, incomplete, and nonnumeric, making analysis with traditional statistical techniques difficult at best. A generic medical data-analysis system called FRID (finding rules in data), which can handle this type of data, is proposed. Incorporating partitioning heuristics, FRID can be used to initiate broad-based exploratory analysis. The system's flexible design also allows a specific search for the probability of a disease given a single symptom. Results from experiments using this system are presented, as well as plans for its future use.