Multiple Machine Learning Models on Credit Card Fraud Detection

Abstract

Nowadays, there is a huge increase in digital financial fraud as a result of the widespread usage of credit cards for online purchases. Therefore, a credit card fraud detection method with high accuracy, minimizing the risk of losing money when the transaction occurs, becomes imperative. The study first processed a synthetic data set, discarding useless features, using one hot encoding to convey categorical information to numerical ones and separating the training and testing datasets. Based on the new formed datasets, the study built three Machine Learning (ML) models for fraud detection, which are the Support Vector Machine (SVM) model, the logistic regression model and the decision tree model, respectively. Next, the study implemented on training these three models by fitting the model on training datasets and predicting the test data in testing datasets. Lastly, the heatmap named confusion matrix was provided to visualize the outcomes, indicating the accuracy of each model. Also, the study uses another four performance measurement scores which are Area under the Receiver Operating Characteristic Curve (ROC AUC score), accuracy classification score, balanced F1 score and precision score for evaluating models. Comparing these three models based on the confusion matrix and the four accuracy scores, the study explored that the decision tree model can achieve the best performance compared to other models in predicting fraud.