Skin cancer classification using novel fairness based federated learning algorithm

Numerous skin conditions fall under the category of dermatological diseases, which make proper diagnosis and treatment planning difficult. Our research centres on tackling these obstacles within the framework of federated learning, a decentralized approach to machine learning. We provide a unique strategy that combines class-weighting strategies to reduce the negative effects of different data distributions among decentralized clients by leveraging the federated average algorithm. We assessed the effectiveness of our approach using the Fitzpatrick 17k dataset, an extensive collection encompasses a wide range of skin conditions. With its realistic representation of dermatological diagnosis scenarios, the dataset provides a solid foundation for training and testing federated learning models. One of the main issues driving our research is the ubiquitous problem of class imbalance within federated learning. When client data distributions are uneven, class imbalance can result in biased model predictions and subpar performance. To solve this issue and enhance model performance, we have incorporated class-weighting approaches into the federated average architecture. We show through thorough experimentation that our strategy is useful for improving federated learning models’ learning performance. Our methodology presents a possible solution to the class imbalance issue in federated learning situations by reducing bias and increasing prediction accuracy. Our study further emphasizes the significance of iterative refinement methods for optimizing federated average weights and fine-tuning model parameters. The results of our study show that the model performance has improved significantly, with an average accuracy of almost 92% across all categories. These results highlight our classification model’s potential usefulness for dermatological diagnosis and treatment planning in clinical settings. Furthermore, this study contributes valuable insights into the application of federated learning for dermatological disease classification, paving the way for future advancements in addressing key challenges such as data privacy, distribution heterogeneity, and model fairness in medical imaging.