Blockchain-aware federated anomaly detection scheme for multivariate data

Real-time sensors in airports, power plants, intelligent manufacturing, and healthcare systems make multivariate time-series anomaly detection more critical. Two significant obstacles remain. First, data organisations isolate sensitive data on islands and train high-performance anomaly detection models to preserve privacy and security. Data organisations have statistical heterogeneity. A unified anomaly detection methodology fails with personalised data. Blochchain-aware federated anomaly detection framework (BcFad) for multivariate time series data. BcFad uses the federated learning architecture to aggregate data while respecting privacy and fine-tuning a reasonably personalised model. BcFad improves F1 scores by 6.9% relative to the baseline technique in NASA spacecraft dataset experiments.