Classification conducting knowledge acquisition by an evolutionary robust GRBF-NN model

Diverse machine learning methods have been successfully used to discover classifying rule among classification data. Sometimes, executing a decision may however indirectly alter feature values of the classified objects, further influencing their classes under the discovered classifying rule. Actually, mining such classification conducting knowledge (CC-knowledge) hidden under the decision from related data can be very helpful to future decision-makings. Hence, this paper proposes an evolutionary robust GRBF-NN model to imitate the mathematical mapping between the feature values before and after executing the decision. A dual-loop nested robust training (DNRT) method is correspondingly developed to determine the weights and parameters using M-AdaBoost and NSGAII respectively in the inner and outer loop. Its remarkable merit is that it considers the classification information by integrating the given classifying rule into both training loops, ensuring the reasonability of prediction. In order to enhance the model’s generalization, the outer loop defines a regularized term and regards it as another optimizing objective of NSGAII besides the training error. Finally, several datasets are employed to verify the effectiveness of the proposed method for CC-knowledge acquisition.