Abstract
Federated learning (FL) has become more popular in the area of machine learning for protecting data privacy, its unique distributed data processing characteristics have garnered widespread attention. However, the implementation of FL faces many challenges, it can be difficult has to decide on a compromise between model security, data privacy, and system efficiency, often requiring the give up of efficiency for privacy, traceability, interpretability, and security. In this paper, privacy-preserving federated incremental learning blockchain-optimized explainable artificial intelligence (PPFILB-OXAI) leveraging the benefits of Blockchain, Federated Incremental Learning (FIL), and explainable artificial intelligence (XAI) with optimization. Chaotic Bobcat Optimization Algorithm (CBOA) is introduced to XAI for selecting most important features from the dataset. The CBOA mimics the instinctive behaviors of wild bobcats, incorporating a chaotic operator to randomly generate the population during the selection phase. It is inspired by the bobcat's hunting tactics, particularly the approach and pursuit of prey. Throughout the algorithm iterations, the most optimal feature solution is gradually identified. The FIL algorithm is capable of adapting to increasing resources in real-time without the need for retraining, all while extracting meaningful patterns from the collective client side data. Meanwhile, Blockchain technology makes it possible to handle medical data securely and transparently, and XAI improves the clarity and understanding of model decisions. To coordinate client privacy protection, PPFILB-OXAI integrates the blockchain process, FIL, and privacy approach. It then uses an aggregate to filter out aberrant models. Lastly, Entropy Deep Belief Network (EDBN) has shown the ability to classify and identify attacks. PPFBXAIO provides the best performance on a breast cancer wisconsin and heart disease in terms of precision, recall, f-measure, accuracy, loss, latency, and throughput. Heart disease, the precision, recall, f-measure, and accuracy of the suggested system are 94.87%, 96.73%, 95.79%, and 95.71%, respectively. The precision, recall, f-measure, and accuracy of the suggested method for breast cancer wisconsin are 97.13%, 97.70%, 97.41%, and 96.84%, respectively.