Application of the Genetic Algorithm to the Rule Extraction Problem
| Autor: | Dounia Yedjour |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Artificial neural network
business.industry Computer science media_common.quotation_subject Evolutionary algorithm Fidelity Machine learning computer.software_genre Set (abstract data type) ComputingMethodologies_PATTERNRECOGNITION Black box Genetic algorithm Extraction methods Artificial intelligence business computer media_common Test data |
| Zdroj: | Artificial Intelligence and Renewables Towards an Energy Transition ISBN: 9783030638450 |
| DOI: | 10.1007/978-3-030-63846-7_57 |
| Popis: | Artificial Neural Networks are considered as a black box. They are unable to explain its classification decision. Several rule extraction algorithms from trained neural networks have been developed to overcome this problem. The aim is to have a set of rules to explain how ANNs solves a given problem. A global rule extraction algorithm from trained neural networks, based on evolutionary algorithms is presented. The extracted rules are evaluated from three criteria: fidelity, accuracy and comprehensibility. The fidelity indicates how the extracted rules mimic the decision of the trained neural networks. The accuracy is calculated from dataset, it indicates the ability of the rules to satisfy the test data. The comprehensibility designates the number of the extracted rules. The proposed method is evaluated on 03 UCI datasets. A tradeoff between the accuracy, the fidelity and the comprehensibility has been showed. The results of these experiments are presented and compared with existing rule extraction methods. Our proposal achieves a best accuracy and comprehensibility over breast cancer dataset. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|