DECISION TREE BASED MACHINE LEARNING ALGORITHMS FOR FRAUD DETECTION SYSTEM

• Main Author: Rifda Hayati, Adiyanti
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/39112
• Institution: Institut Teknologi Bandung

One method of digital transactions that is rampant today besides e-banking and m-banking is mobile money or cellular money. But the use of mobile money faces various challenges, particularly with regard to security. Various types of fraudulent constitute a separate risk for users and service providers of financial transactions. Research on these issues is carried out in five stages, understanding business needs, data understanding, data preprocessing, modeling and evaluation. Decision tree based algorithms that used in this research are CART, Random Forest and XGBoost. These model are applied to PaySim dataset and evaluated using four different metrics, there are recall, specificity, precision and F1 score, with recall and specificity metrics as the main standard. The results of this research indicate that when applied to PaySim test data, the XGBoost algorithm provides better results than CART and Random Forest algorithms. However, when applied to the backtesting data, the all tree algorithms provide poor results with low specificity value.