Multi-objective Optimization of Meta-learning Scheme for Context-based Fault Detection

The paper deals with the problem of performance optimization of a meta-learning scheme for context-based fault detection. The context-based ensemble classifier is proposed to increase the performance of the fault diagnosis system. The most important problem to solve in this approach is to find optimal structures as well as optimal values of behavioral parameters of component classifiers. This problem has been elaborated as a multi-objective optimization task taking into account different objectives obtained from a confusion matrix. It was decided to make use of the NSGA-II algorithm in order to search for the optimal solution. A case study is based on the laboratory stand for simulation of hydraulic industrial processes. Common machine learning methods such as decision tree, naive Bayes, Bayesian network and k-nearest neighbors were taken into account in the meta-learning scheme for context-based fault detection. The obtained results prove that the proposed approach has practical relevance.