Implementation of hybrid sampling technique for predicting active compound and protein interaction in unbalanced dataset
Wisnu Ananta Kusuma, Alya Rahmi, Rudi Heryanto · IOP Conference Series Earth and Environmental Science · 2019
Abstract Indonesia Jamu Herbs (Ijah) web server aims to predict Jamu efficacy based on interaction between active compound and disease’s protein. However, the interaction between compound and protein data is unbalance since there are many unknown interactions between active compounds and protein target. Thus, the prediction result is still not optimal. In this research, the hybrid sampling technique, combining complementary fuzzy support vector machine (CMTFSVM) and synthetic minority oversampling technique (SMOTE) was used to handle imbalanced data interaction between active compound and protein for Ijah, web server to predict candidate Jamu formula for certain disease. Performance was measured using geometric mean (Gmean), area under curve (AUC), and accuracy. The evaluation results showed that the hybrid sampling technique could increase the instance of minority class three times. Moreover, the prediction model could obtain the value of 0.8346, 0.6812, and 0.5319 for accuracy, Gmean, and AUC, respectively.