Multi-Objective PSO-fuzzy Optimization Approach to Improve Interpretability and Accuracy in Medical Data
Alwatben Batoul Rashed A, Hazlina Hamdan, Md. Nasir Sulaiman, Nurfadhlina Mohd Sharef, Razali Yaakob · International Journal of Engineering & Technology · 2018
Today, Decision Support Systems (DSS) plays a significant role in a medical and healthcare domain. Designing an Automatic FuzzyRule-based Classification Systems (FRBCSs) is considered as optimization problem associated to a result of high interpretability andaccuracy. Interpretability and accuracy are the two main objectives to be improved in the optimization measurement of FRBCSs. However, improving these objectives is found to be difficult in most of the existing systems due to the conflicting issues between accuracyand interpretability. In this work, we proposed an approach that can effectively handle accuracy- interpretability trade-off in constructingFRBCSs. We designed automated FRBCSs in the form of Multi-objective Particle Swarm Optimization with Crowding Distance. In theapproach, there will be a collection of solutions to FRBCSs that deem best global minimum or global maximum with respect to interpretability and accuracy. Our method is evaluated on a popular benchmark data sets being used in a medical domain for evaluations. Thesedatasets are Liver Disorders (BUPA), Pima Indians Diabetes and Thyroid Disease (New Thyroid). The result obtained shows that theproposed method yields an optimum solution in minimizing the trade-off between accuracy and interpretability. Moreover, the result ofthe comparison shows that our approach outperforms the alternate techniques in terms of accuracy of FRBCSs and also exhibits goodresult in terms of interpretability objective.