Feasibility Studies of Quality of Knowledge Mined from Multiple Secondary Sources

摘要

A new algorithm, aimed at evaluation of learning models developed by inductive machine learning from examples, was developed and tested. The evaluated learning models are treated here as secondary sources of information, giving an indirect insight into the primary source of information, being the standard decision table. It was found, that after application of the developed algorithm, the number of rules generated for an investigated decision table, was reduced from 45 (on average, depending on the method of quantization of continuous attributes) to 4, without increasing the error rate.