Main Article Content
The act of fraudulent credit card transactions has been increased over the past recent years, as the era of digitization hits our day-to-day life, with people getting more involved in online banking and online transaction system.
Machine learning algorithms have played a significant role in detection of credit card frauds. However, the unbalanced nature of the real-life datasets causes the traditional classification algorithms to perform low in detection of credit card fraud.
In this work, a cost-sensitive weighted random forest algorithm has been proposed for effective credit card fraud detection. A cost-function has been defined in the training phase of each tree, in bagging which emphasizes assigning more weight to the minority instances during training. The trees are ranked according to their predictive ability of the minority class instances. The proposed work has been compared with two existing random-forest based techniques for two binary credit card datasets. The efficiency of the model has been evaluated in terms G-mean, F-measure and AUC values. The experimental results have established the proficiency of the proposed model, than the existing ones