Fuzzy classifier based on clustering with pairs of ε-hyperballs and its application to support fetal state assessment. (15th March 2019)

• Record Type: Journal Article
• Authors: Jezewski, Michal; Czabanski, Robert; Leski, Jacek M.; Jezewski, Janusz
• Is Part Of: Expert systems with applications. Volume 118(2019)
• Page Start: 109
• Page End: 126
• Publication Date: 2019-03-15
• Subjects: Fuzzy classifier -- Fuzzy rule extraction -- Fuzzy clustering -- ε-Insensitivity -- Fetal monitoring; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2018.09.030
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: Pairs of ε-hyperballs improve the effectiveness of fuzzy classification. Clustering with pairs of prototypes efficiently supports automated fetal state assessment. Two-step cardiotocogram analysis increases sensitivity of fetal diagnosis. Abstract: Objective: In this study we propose a fuzzy classifier whose rule antecedents are determined based on the new approach to Clustering with Pairs of Prototypes (CPP). After demonstrating the high generalization ability of the classifier for six various benchmark datasets, a particular emphasis was placed on the application to support fetal state assessment based on the classification of cardiotocographic (CTG) signals. Methods: The CPP is a solution aimed at increasing the performance of fuzzy classifiers by introducing additional prototypes to those obtained using a given basal clustering method. The CPP improvement was achieved by applying the Fuzzy Clustering with ε-Hyperballs (FCεH) as basal clustering, as well as a new ant algorithm-based method of searching for pairs of prototypes. Results: The results were compared with three reference methods: the Lagrangian SVM with the Gaussian kernel function, and the same fuzzy classifier, but using the antecedents determined with the fuzzy c -means and the fuzzy ( c + p ) -means clustering. In case of five out of six benchmark datasets as well as for the CTG signals classification problem we achieved the highest generalization ability measured with the classification accuracy (benchmark data) and the classification quality index defined as geometric mean of sensitivity and specificity (CTG signals). Conclusions: The results of the numerical experiments showed high accuracy of the CPP-based fuzzy classifier when assessing various types of data. Moreover, the two-step classification of the CTG signals based on the proposed method allows for the efficient signal evaluation aiming to support the automated fetal state assessment. Significance and main impact: The most significant feature of the proposed method is the high generalization ability being the result of the ε-insensitive learning (FCεH clustering), while maintaining the possibility of interpreting the learning outcomes thanks to the linguistic representation of the knowledge in the form of fuzzy conditional (if-then) rules. Therefore, we believe that this solution will have a positive impact on other studies on intelligent systems.