Utilizing Machine Learning Techniques to Identify Anomalies in Financial Transactions for Laos Banking

Recently, rapid technological advancements have sparked the growth and widespread adoption of new financial industries. This includes internet-based services such as applications, credit cards, ATMs, online shopping payments, and online money transfers. As these systems become more sophisticated, the risks and threats also increase significantly. This research aims to explore ways to protect the assets of customers and banks in Laos from falling victim to individuals with malicious intent who attempt to fraudulently steal customer assets online. New technologies, like creating anomaly detection models for financial transactions, safeguard customer and bank assets by filtering out suspicious activities before confirming the actual transaction. If the system detects an anomaly in a transaction, it will alert the account holder and ask for confirmation before proceeding with the transaction. We trained the anomaly detection model on a dataset of over 12 million records, utilizing techniques like isolation forest (IF), local outlier factor (LOF), and autoencoder (AE). We assessed the effectiveness of these three models, obtaining ROC scores of IF = 0.54, LOF = 0.50, and AE = 0.92, as well as training times (in minutes) of IF = 3.37, LOF = 17.42, and AE = 8.57. The AE model was shown to be the most successful in detecting anomalies, indicating that it might be used as an API to improve the monitoring of financial transactions. Additionally, the lack of labeled data, which is typical in fraud detection tasks, was handled via the use of unsupervised learning. But because training AE needed a significant amount of computing power, the study was constrained, which would have affected its viability for real-time applications.