Induced decision trees for time- and cost-sensitive data
Date of Award
Doctor of Philosophy (PhD)
Electrical Engineering and Computer Science
expert systems, knowledge induction, time sensitivity
Electrical and Computer Engineering
Rule induction serves as an alternative knowledge acquisition method in the development of expert systems. Time- and cost-sensitive data is present in many common induction applications, such as predicting the bankruptcy of corporations and approving their credit. Data that affects these decisions changes with time. Present rule induction methods do not account for the varying availability of data, and research on the cost of data in rule induction is contradictory. Utilizing the presence of these cost factors can result in a more valuable decision support tool, one that can classify cases from partial information. Making such early decisions trades reduced accuracy for the benefits of early classification. Previous research has selected attributes and stopped tree development on the basis of cost. This research identifies some weaknesses in the previous research and offers modifications to the methodology The result is a set of guidelines for constructing more cost-effective induced decision trees. Experiments validated the usefulness of cost pruning and identified conditions in which a given cost selection measure may be superior. The data for the experiments came from two applications: approving loans (an expert's decisions) and predicting a hospital patient's length of stay (actual outcomes). In the hospital application, the goal is to identify the early characteristics that predict excessive length of stay. The medical data accumulates over time--patient history data is followed by data that arrives during the hospitalization period. The consideration of time as a factor in the cost of a decision tree is a new element in induction research.
Surface provides description only. Full text is available to ProQuest subscribers. Ask your Librarian for assistance.
Murphy, Catherine Kuenz, "Induced decision trees for time- and cost-sensitive data" (1997). Electrical Engineering and Computer Science - Dissertations. 233.