Optimization of Fuzzy Support Vector Machine (FSVM) Model in Multiple Metric Spaces

Fuzzy membership function was introduced into the Support Vector Machine (SVM) resulting in modifications. Selecting the correct membership function is an important step in the Fuzzy Support Vector Machine (FSVM) method. One of the general criteria for selecting fuzzy membership is determined by the distance between a point and its fixed center category. This study aims to develop the SVM method into Fuzzy SVM (FSVM) with several distance functions that are applied to the Early Stage Diabetes data which collects 520 data. The distance functions used include Euclid, Canberra distance, Minkowski distance, Chebyshev distance, Minkowski Chebyshev distance, and Bray-Curtis distance where this distance function is used to determine the best distance that can be seen from the results of accuracy, specificity, g-means which is best for viewing diabetes risk. The results of this comparison show that the FSVM method with several distance functions is more than the SVM method. Where the FSVM method at the Canberra distance with a penalty value of C = 2
⁵
is the best distance to see the risk of diabetes, based on the results of specificity = 100%, g-means = 86.91%, and accuracy = 85.26% is superior to the SVM method at the penalty value C = 2
¹⁰
with specificity = 69.36%, g-means = 77.31%, and accuracy = 79.49%. Although the FSVM method produces an evaluation value at sensitivity = 75.53%, it is lower than the SVM method with a sensitivity value = 86.17%.