Detecting cyberattacks using anomaly detection in industrial control systems: A Federated Learning approach

Light-weight federated learning-based anomaly detection for time-series data in industrial control systems

“…Furthermore, the authors of [18] did not account for the model training time, which is the major drawback of LSTM and GRU. Work [19] proposed a cyberattacks detection mechanism using the combination of Variational Autoencoder (VAE) and LSTM. Although the detection performance is considerably high, the model in [19] faces difficulty in terms of running time since the LSTM block requires a long training time.…”

“…Work [19] proposed a cyberattacks detection mechanism using the combination of Variational Autoencoder (VAE) and LSTM. Although the detection performance is considerably high, the model in [19] faces difficulty in terms of running time since the LSTM block requires a long training time. Liu et al [20] presented an attention CNN-LSTM model within a FL framework for anomaly detection in IIoT edge devices.…”

“…However, ATRAF in both modes is proven to detect anomalies efficiently. In this experiment, the detection performance of FATRAF is compared with a recent detection work for ICS that applies Federated Learning [19] -Federated Learning called FL-VAE-LSTM. As Table 4 shows, FATRAF improves the detection performance in all metrics: Precision, Recall, F1score with both of the ICS data sets: Power Demand and Gas Pipeline.…”

A review of Federated Learning in Intrusion Detection Systems for IoT

“…[91], [92], [93], [94] [95], [96], [97], [98] [99], [100], [101], [102], [103] [104], [105], [106], [107], [108] Fig. 2: Existing Deep Learning Federated Intrusion Detection Systems by model architecture.…”

“…Huong et al [92] proposed a cyberattack anomaly detection system for Industrial Internet of Things (IIoT) systems. In order to detect anomalies in time series data, they use a model architecture composed of an VAE-Encoder, an LSTM unit and an VAE-Decoder.…”