EFFECTIVE ANOMALY DETECTION OF E-WALLET TRANSACTIONS USING DEEP LEARNING AND SPATIO- TEMPORAL ANALYSIS

Electronic wallet transactions are increasingly popular and becoming a part of everyday life. However, as the number of transactions increases, the risk of anomalies and fraud also increases. This study aims to develop an anomaly detection model in e-wallet transactions using the Recurrent Neural Network (RNN) model, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) and spatio-temporal analysis.
The transaction data used in this study comes from the Mony platform and includes various information such as transaction IDs, timestamps, transaction amounts, geographical coordinates (latitude and longitude), as well as sender and recipient account information. The steps in the data processing process include: first, data pre-processing to correct and fill in the missing values; second, the extraction of spatio-temporal features to calculate the distance, time difference, speed of movement between transactions, and calculate the habit of the number of transactions; and third, data labeling to identify normal and anomalous transactions.
The built Deep Learning model consists of two layers with 256 and 128 units respectively, equipped with a dropout layer to prevent overfitting. The data were trained using the SMOTE oversampling technique to handle data imbalances and then divided into training, validation, and testing sets. Model evaluation was carried out using accuracy, precision, recall, and F1-score metrics.
The evaluation results show that this LSTM model has the best performance with a test accuracy of 94.64%, precision of 94.81%, recall of 94.64%, and F1-score of 94.62%. The GRU model showed a test accuracy of 94.46%, precision of 94.86%, recall of 94.46%, and F1-score of 94.39%. The RNN model has lower performance with a test accuracy of 91.78%, precision of 91.74%, recall of 91.78%, and F1-score of 91.76%. This shows that the LSTM model developed has excellent ability to detect anomalies in e-wallet transactions.

Keywords: anomaly detection, e-wallet transactions, Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), spatial- temporal analysis, deep learning.