Credit Card Fraud Detection via Machine Learning

Credit card fraud has recently been a more severe problem since fraudsters are continuously developing more sophisticated fraud techniques. In this work, we have developed a software-based machine learning (ML) system for detecting credit card fraud. Data pre-processing was the first step at which Pearson correlation coefficient was used to decide which features are worth using in our model. Then, k-folds cross validation was used to evaluate the performance of different classifiers: k-Nearest Neighbor (k-NN), Naïve Bayes, Support Vector Machine (SVM), Bagging, Random Forest and Multilayer Perceptron (MLP). To overcome the imbalanced data issue, synthetic minority oversampling technique (SMOTE) was applied to the dataset to generate more fraud data points. To evaluate model's performance, Precision, recall, accuracy and F1 score evaluation metrics were used. With feature correlation of 0.1, SVM had the highest recall score: 88.55%. Also, when implementing SMOTE, the k-NN classifier showed the highest F1 score and precision.