Abstract
In this paper, efficiency of Support Vector Machine (SVM) and Artificial Neural Network (ANN) is analyzed using unbalanced dataset. The dataset analyzed in this study is obtained from COIL Challenge'2000 and it is highly unbalanced with 94% good customers' data and 6% bad or fraud customers' data. We employed balancing techniques and SMOTE to bring the balance in the data and analysis is carried out. We employed (1) Under-sampling, (2) Over-sampling and (3) Synthetic Minority Oversampling Technic* (SMOTE) for balancing the dataset. Since identifying fraudulent cases is paramount from the business perspective, management accords higher priority on sensitivity only. Therefore considering sensitivity alone, we observed that SVM outperformed with original unbalanced data. It is also observed that NN performed better with balanced data compared to it performance using unbalanced data.