Deep Autoencoder Ensembles for Anomaly Detection on Blockchain

Distributed Ledger technologies are becoming a standard for the management of online transactions, mainly due to their capability to ensure data privacy, trustworthiness and security. Still, they are not immune to security issues, as witnessed by recent successful cyber-attacks. Under a statistical perspective, attacks can be characterized as anomalous observations concerning the underlying activity. In this work, we propose an Ensemble Deep Learning approach to detect deviant behaviors on Blockchain where the base learner, an encoder-decoder model, is strengthened by iteratively learning and aggregating multiple instances, to compute an outlier score for each observation. Our experiments on historical logs of the Ethereum Classic network and synthetic data prove the capability of our model to effectively detect cyber-attacks.