Comparative Evaluation of Machine Learning Algorithms with Parameter Optimization and Feature Elimination for Fraud Detection

Machine Learning (ML) algorithms could be performed efficiently for lots of areas range from education, medicine, defense industry to consumer applications and finance. Especially data classification in finance area has become a striking part of literature recently. Especially, data classification in finance area has become a striking part of literature recently. On this issue, 5 well known ML algorithms which are Logistic Regression Classifier, Decision Tree Classifier, AdaBoost Classifier, K Nearest Neighbor (KNN) Classifier and Support Vector Classifier (SVC) are evaluated by using "Credit Card Fraud Detection" dataset to deal with Fraud and Non-Fraud classification. After all algorithms are performed, it is seen that SVC has the best f1 and precision-recall scores. Moreover, ANOVA is more useful strategy to eliminate irrelevant features compared to Mutual Information method for the dataset. Optimization of model parameters are also critical factor improving classification performance.