Rule extraction from trained neural networks using genetic algorithms
Artificial Neural Networks (ANN) are generally used as universal optimizers. Given a set of input and output values of a specific problem, no matter how complex the problem is, an ANN can easily learn the relationship between the input and output domain, and therefore can generalize over a set of unseen values prosperously. The knowledge acquired by an ANN during its learning is encoded in two aspects: i) the network architecture itself (i.e., the number of hidden units); ii) the network parameters (i.e., weight vectors). Generalization is managed using this encoded knowledge, thus concepts learned by an ANN are hidden in its architecture and weights. Despite the effective generalization capability, ANNs bear the inability to provide an accurate and comprehensible explanation of what has been learned. One approach to overcome such a drawback is to extract symbolic rules from trained ANNs which best describe the internal behavior of the networks. The rules are generally expressed as Disjunctive Normal Form (DNF) in IF-THEN rule structure, thereby allowing multi-layer perceptrons to be used suitably to serve this purpose. In our approach, we use a genetic algorithm to optimize an ANN topology, and then use decompositional approach to extract rules in disjunctive normal form from the optimized and trained ANN. The lay-out of the paper is as follows. Section 2 describes our approach in detail. Section 3 outlines the test case used in this study: the three monk's problems. Section 4 presents the simulation results, and finally Section 5 concludes this study.
- A New Approach to the Extraction of ANN Rules and to Their Generalization Capacity Through GP
- Discovering the Mysteries of Neural Networks
- EXPLANATORY RULE EXTRACTION BASED ON THE TRAINED NEURAL NETWORK AND THE GENETIC PROGRAMMING
- Artificial Intelligence and Soft Computing - ICAISC 2004
- scientific article; zbMATH DE number 2087697
- Greedy rule generation from discrete data and its use in neural network rule extraction
- Review of research in the field of developing methods to extract rules from artificial neural networks
- Generating rules with predicates, terms and variables from the pruned neural networks
- A search technique for rule extraction from trained neural networks
- Artificial Intelligence and Soft Computing - ICAISC 2004
- scientific article; zbMATH DE number 1928724 (Why is no real title available?)
- scientific article; zbMATH DE number 2087697 (Why is no real title available?)
- A New Approach to the Extraction of ANN Rules and to Their Generalization Capacity Through GP
- EXPLANATORY RULE EXTRACTION BASED ON THE TRAINED NEURAL NETWORK AND THE GENETIC PROGRAMMING
- A dual-objective evolutionary algorithm for rules extraction in data mining
- Neural network explanation using inversion
This page was built for publication: Rule extraction from trained neural networks using genetic algorithms
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4377233)
