A Neural Networks Based Approach for Fast Mining Characteristic Rules

摘要

Data mining is about extracting hidden information from a large data set. One task of data mining is to describe the characteristics of the data set using attributes in the form of rules. This paper aims to develop a neural networks based framework for the fast mining of characteristic rules. The idea is to first use the Kohonen map to cluster the data set into groups with common similar features. Then use a set of single-layer supervised neural networks to model each of the groups so that the significant attributes characterizing the data set can be extracted. An incremental algorithm combining these two steps is proposed to derive the characteristic rules for the data set with nonlinear relations. The framework is tested using a large size problem of forensic data of heart patients. Its effectiveness is demonstrated.