FedShield: Blockchain and Federated Learning Based Collaborative Framework for Windows Malware Detection in Smart Applications

As new and upcoming technological advancements emerge, the threat of malware will continue to diversify and increase. Malware poses a serious threat to privacy and security of critical data. Some malware do not collect data but use the device's resources for crypto-mining and other activities. Thus, new technology for evading and detecting malware continues to grow and advance. Artificial Intelligence (AI) is one such field that has helped tackle this problem with great precision. Most of the existing solutions use Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and algorithms to make more accurate predictions. However, their solutions are not optimized due to generalized training and higher latency due to large model structures. Inspired by the aforementioned challenge, this paper proposes a distributed learning approach called FedShield to solve the said problem using a Windows Malware dataset. Federated Learning (FL) is one such algorithm that helps to train models distributively, ensuring data security, privacy, scalability, and diversification. Each client model has an ANN that communicates with the global model to update its model via a blockchain layer. The FL model achieves an accuracy of 96 % with 13 clients on the unseen data. Furthermore, the blockchain layer also stores the malicious files in order to make them tamper-proof and secure. The proposed FedShield system is evaluated by comparing it with pre-existing models. The transaction and execution costs in the blockchain for each function are recorded. This approach can help various anti-malware softwares to improve their products.