Influence of fuzzy index parameter on FSVM classifier performance

Abstract

Support vector machine (SVM), a machine learning algorithm used extensively for pattern analysis and recognition, is found sensitive to outliers and noise. Fuzzy support vector machine (FSVM) has been used in many applications as a most prominent technique by researchers to overcome the sensitivity issue faced by SVM, and for its good generalisation performance. In this research, a method to justify the performance of FSVM classifier by showing the influence of fuzzy index m on membership function of the model has been proposed. In the first phase of the study, an algorithm to find the optimal fuzzy index using fuzzy C-means (FCM) and to avoid testing all fuzzy index values on the FSVM model has been proposed. The operational complexity of the model can be reduced by this process. In the second phase of this study a FSVM algorithm to incorporate new membership function is proposed. The model is tested on six different datasets and kernel functions and the kernel with the highest classification accuracy is identified as an efficient kernel for the applied dataset. The experimental results on the chosen fuzzy index have proven that the proposed alternate methodology enhances classifier accuracy compared to other research findings. Hence, the model could be applied to diverse fields of FSVM applications.