A Study on False Positive Reduction in Unbalanced Datasets through SVM Using Cost Based Learning

Fraud Detection in unbalanced Datasets victimization price primarily based Learning uses price complicated Support Vector Machines (CCSVM) for locating the ineligible transactions. SVM could be a binary classification, therefore the transactions square measure labeled either as dishonest or legitimate. To handle the unbalanced dataset, the formula victimization totally different error price for the positive (+) and therefore the negative (-) categories is employed. SVM weight implements price sensitive learning. just like SVM, the weighted SVM is employed to maximize the margin of separation and minimize the classification error. The margin boundary is employed to separate the categories. In CS-SVM totally different weights square measure assigned to the categories. Effective call boundary is learned by adjusting the weights of the various categories. It improves the accuracy of the prediction rate.