ANOMALY DETECTION SYSTEM IN BANKING TRANSACTIONS USING A MACHINE LEARNING MODEL BASED ON GRAPH NEURAL NETWORK

• Main Author: Aziz Bukhori, Hilmi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73912
• Institution: Institut Teknologi Bandung

Graph machine learning and fraud detection systems are growing and popular today. Fraud detection systems have been widely used as a tool to detect potentially fraudulent transactions. Fraud detection systems can be used to determine patterns of transactions that are suspected of being criminal transactions. Graph machine learning development can be implemented in anything that can be represented in graph form. The banking fraud detection system can be implemented in graph form by connecting customers who have made transactions with other customers or customer transactional activities. From the graph that has been formed, predictions will be made so that new transactions can be classified as fraudulent transactions or not by connecting these transactions with the graphs that have been made. The experimental results show that the graph-based fraud detection model produces better performance than the tree-based model, but with a longer inference time. to try to reduce the inference time generated in the graph model, the authors also do sampling with importance sampling method. The results shown, with accuracy that is still higher than the tree-based model, sampling can reduce the value of inference time in the graph model, although it is still at a higher value than the tree-based model. Based on the experimental results, the best AUC Score was produced by the GraphSage model with an AUC of 0.997, but based on the fastest inference time, it was produced by LightGBM from a tree-based model, namely 3.96 ms. With the implementation of importance sampling, the resulting AUC is 0.994 with an inference time of 503 ms.