A Novel Method for Extracting Knowledge from Neural Networks with Evolving SQL Queries

While artificial neural networks (ANNs) are undoubtedly powerful classifiers their results are sometimes treated with suspicion. This is because their decisions are not open to inspection - the knowledge they contain is hidden. In this paper we describe a method for extracting and representing the knowledge within an ANN. Mappings between inputs and output classifications are stored in a table and, for each classification, Structured Query Language (SQL) queries are evolved using a genetic algorithm. Each evolved query is a simple, humanreadable representation of the knowledge used by the ANN to decide on the classification based on the inputs. This method can also be used to show how the knowledge within an ANN develops as it is trained, and can help to identify problems that are particularly hard, or easy, for ANNs to classify.

[1]  Lipo Wang,et al.  Rule extraction using a novel gradient-based method and data dimensionality reduction , 2002, Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN'02 (Cat. No.02CH37290).

[2]  Wlodzislaw Duch,et al.  A new methodology of extraction, optimization and application of crisp and fuzzy logical rules , 2001, IEEE Trans. Neural Networks.

[3]  Bo Yu,et al.  Extract rules from software quality prediction model based on neural network , 2004, 16th IEEE International Conference on Tools with Artificial Intelligence.

[4]  M. Kasiran,et al.  An information framework for a merchant trust agent in electronic commerce , 2002 .

[5]  A. Pazos,et al.  Automatic recurrent ANN rule extraction with genetic programming , 2002, Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN'02 (Cat. No.02CH37290).

[6]  Xin Yao,et al.  Evolving SQL Queries for Data Mining , 2002, IDEAL.

[7]  Lipo Wang,et al.  Rule extraction from an RBF classifier based on class-dependent features , 2002, Proceedings of the 2002 Congress on Evolutionary Computation. CEC'02 (Cat. No.02TH8600).