Elastic version space: a knowledge acquisition method with background knowledge adjustment

Similarity based learning (SBL) is efficient in knowledge acquisition process, and it uses training examples to generate rules and refine them. Training examples collected in the real world are very often constructed with numerical attributes. In order to deal with these training examples, SBL needs background knowledge. Especially segments which specify value ranges of numerical attributes are discussed in the background knowledge. Elastic version space method is proposed here which integrates the version space method with the functions of segments adjustment. By defining the segments structure using margin segments in the background knowledge, the version space method itself adjusts the segments. In consequence, this method expands the region of application of version space. Empirical results applying to the man-power allocation problem are presented which shows that the elastic version space method is an effective SBL in the knowledge acquisition process.