ANOMALY DETECTION IN NEFT TRANSACTIONS

Anomaly detection-a fundamental approach in data mining and machine learning-is based on finding any deviations in data patterns with respect to the expected behavior. This area of study finds application in various fields such as fraud detection, network security, and financial systems, where pinpointing any irregularities is key to safety and integrity. Anomaly detection in NEFT transactions was used to continually monitor the patterns of transactions for fraud or abnormality that would indicate security threats or human error. NEFT transactions represent high-volume data and require real-time access, so it is essential to use robust and powerful algorithms capable of finding minutiae in large datasets. In this study, three popular unsupervised anomaly detection algorithms-Local Outlier Factor (LOF), Isolation Forest, and Autoencoders-have been employed. LOF identifies various anomalous points according to deviations from the local density, thus effectively spotting outliers in transaction data interaction complexities. Isolation Forest does this by recursively isolating the anomalous data points, thus offering great efficiency in dealing with high-dimensional datasets. Autoencoders are a type of neural network with excellent data representation-learning ability; they mark a broad scope to reconstruct normal transaction patterns and flag all atypical patterns as anomalies. This research discusses and evaluates the relative performance of these methods on NEFT transactions, along with the salient features of each model in capturing anomalies within financial datasets. By investigating robust methodologies suitable for NEFT transactions, the study provides a significant boost towards enhancing fraud detection in financial systems. Keywords — Anomaly detection, NEFT transactions, fraud detection, Local Outlier Factor, Isolation Forest, Autoencoders, unsupervised learning, financialdatasets.