Taylor Series‐based Activity and Trust Aware Federated Learning Aggregation for Malware Detection in IoT

Abstract

The Internet of Things (IoT) has been developed progressively because of its effective data transmission. Still, the IoT devices are affected by malware attacks. The malware detection in IoT is complex owing to the varied capacities of devices. The Federated Learning (FL) is utilized for enhancing privacy against malware activities. Hence, this paper develops the Taylor Series‐enabled Activity and Trust Aware Federated Learning Aggregation (TaTFedLA) for malware detection. The client and the server are the components in FL, in which steps like data acquisition, preprocessing, feature fusion, and malware detection are performed in the training. The Min‐Max normalization is utilized for preprocessing. Deep Kronecker Network (DKN) with a non‐correlation similarity is utilized in feature fusion. The Fractional Lotus Effect Optimization‐based Deep Belief Neural Network Fused SpinalNet (FLEO_DBNFSpinalNet) using a FL approach is employed for malicious detection. After evaluating the trained model, the server controls model aggregation. Hence, trust is established between the server and the IoT device, and aggregation is done using the Taylor series concept. Moreover, the TaTFedLA‐based malware detection achieves the accuracy, Mean Average Precision, False Positive Rate (FPR), loss, Mean Square Error (MSE), Root MSE (RMSE), bandwidth, energy, precision, recall and F‐measure, of 94.99%, 94.25%, 0.121, 5.008, 0.107, 0.326, 36.156 Mbps, 484.318 J, 96.14%, 87.90%, and 91.84%.