ANOMALY DETECTION: IMPROVING DIGITAL TRANSACTION FRAUD DETECTION WITH A HYBRID MODEL OF ISOLATION FOREST AND LONG SHORT-TERM MEMORY (LSTM)

Abstract

Financial transactions in the digital era are increasingly shifting to cashless methods due to their efficiency, but this has actually increased the risk of financial crime. This research aims to develop a more effective anomaly detection model for identifying suspicious transactions in digital transaction systems. This study presents a novel hybrid approach that combines the Isolation Forest Algorithm and Long Short-Term Memory (LSTM) to identify fraud. Isolation Forest was used to detect transactions that deviate from the normal user profile, while LSTM was used to detect various types of fraud. The results showed that out of 10,000 transaction data, the anomaly detection rate reached approximately 1.87% of the total transactions, with accuracy (100%), precision (99%), recall (100%), and F1-score (100%). The evaluation showed that this model was able to detect suspicious transactions with a very low error rate. The study concluded that this novelty is expected to contribute to the development of more accurate and efficient fraud detection models.