BLOCKCHAIN ENABLED FEDERATED LEARNING BASED PRIVACY PRESERVED HEALTHCARE CLASSIFICATION USING FRACTIONAL FOOTBALL OPTIMIZATION ALGORITHM DRIVEN DEEP LEARNING

Abstract

Privacy preserved healthcare classification involves classifying the medical data, ensuring security of sensitive patient information. Blockchain enabled Federated Learning (FL) is emerging paradigm with data privacy, and security across decentralized healthcare systems. However, challenges such as scalability limitations and increased latency are observed. To eradicate these issues, Fractional Football Optimization Algorithm based Deep High-order Attention Neural Network (FFbOA_DHA-Net) is proposed for healthcare classification. Firstly, local training is done at each local node using local data. In training model, the input image is acquired initially and then pre-processing is done by midpoint filter. Further, lesion segmentation is done using TBConvL-Net, and next feature extraction is performed. Finally, healthcare classification is done using DHA-Net, that is trained by FFbOA. Here, FFbOA is designed by merging Football Optimization Algorithm (FbOA) and Fractional Calculus (FC). Moreover, weights from various local training model are aggregated and averaged in global server. Next, global model is applied on every local node based on averaged weights. Furthermore, performance of FFbOA_DHA-Net is assessed with metrics like, Mean Squared Error (MSE), Mean Average Precision (MAP), F1-measure, Root Mean Squared Error (RMSE), and loss, that attained superior values of 0.156, 96.81%, 95.75%, 0.395, and 0.034.