Classical and Deep Learning Classifiers for Anomaly Detection

Abstract

The recent years have seen a heightened interest in the realm of machine learning (ML) for cyber crime detection. The ML approaches use to identify the hidden patterns from the data where the behavior does not comply with usual pattern are called anomaly detection techniques. In this paper we have proposed a 10-layer deep Variational Auto-Encoder (VAE) neural network approach and perform a detailed comparison with different classical classifiers, namely Decision Tree, Support Vector Machine and Ensemble Classifiers. We have employed these classifiers on the credit card transaction anomaly dataset (available online). In conclusion the paper provides the evaluation of each classifier (supervised/unsupervised) through ROC curve, F1-score and confusion matrices.