Authors: Ali Mohammadi Shanghooshabad; Mohammad Saniee Abadeh
Addresses: Faculty of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran ' Faculty of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran
Abstract: Interpretability of extracted fuzzy rules from medical datasets is one of the most important problems in the medical domain. Often, people consider interpretability as Sum of Rules Lengths (SORL) and Number of Rules (NOR) in a rule-based domain, but an important issue which is usually ignored is the variance of the final result (Accuracy, SORL, NOR). This paper considers the variances of accuracy, SORL and NOR as essential interpretability measures. This paper proposes a parallel swarm-based framework to generate multi-objective fuzzy rule-based systems on three medical datasets that decreases the variances of Accuracy, SORL and NOR and simultaneously improves the final Accuracy, SORL and NOR values. Results show that we have been successful in improving the two objectives that were negatively correlated and accordingly we have been successful in generating robust fuzzy rule-based systems.
Keywords: medical knowledge discovery; particle swarm optimisation; PSO; fuzzy rule extraction; multi-objective optimisation; clustering; evolutionary fuzzy systems; robust knowledge discovery; learning classifiers; parallel framework; interpretability; medical data mining; fuzzy logic; bioinformatics.
International Journal of Data Mining and Bioinformatics, 2016 Vol.14 No.1, pp.22 - 39
Available online: 30 Nov 2015Full-text access for editors Access for subscribers Purchase this article Comment on this article