Abstraction of Long-Term Changed Tests in Mining Hepatitis Data

For data mining from time related data, how to deal with data sequences is one of the most important problems. We developed a temporal abstraction (TA) method for mining the hepatitis dataset provided as a common challenge by Chiba university hospital. TA transforms original data sequences into the categorical data and enables to apply various learning methods to transformed data. This paper focuses on temporal abstraction for long-term changed tests with the introduced notion of “changes of state” and an algorithm for extracting them.