Detecting Data Poisoning Attacks in Federated Learning for Healthcare Applications Using Deep Learning

Abstract

This work presents a novel method for securing federated learning in healthcare applications, focusing on skin cancer classification. The suggested solution detects and mitigates data poisoning attacks using deep learning and CNN architecture, specifically VGG16. In a federated learning architecture with ten healthcare institutions, the approach ensures collaborative model training while protecting sensitive medical data. Data is meticulously prepared and preprocessed using the Skin Cancer MNIST: HAM10000 dataset. The federated learning approach uses VGG16's powerful feature extraction to classify skin cancer. A robust strategy for spotting data poisoning threats in federated learning is presented in the study. Outlier detection techniques and strict criteria flag and
evaluate problematic model modifications. Performance evaluation proves the model's accuracy, privacy, and data
poisoning resilience. This research presents federated learning-based skin cancer categorization for healthcare
applications that is secure and accurate. The suggested approach improves healthcare diagnostics and emphasizes
data security and privacy in federated learning settings by tackling data poisoning attacks.