With the rapidly growing daily data and the increasing awareness of customers' privacy, the public applies Federated Learning to solve the data island problem in model training. Deep learning has relatively high requirements on edge computing devices, and XGBoost, which has performed well in both industrial practice and Kaggle competition, becomes a good choice. However, as many clients participate in Federated Learning, statistical heterogeneity occurs due to global and local class imbalance. Combined with focal loss, our proposed new algorithm alleviates the above phenomena through comprehensive scenario experiments. In addition, we show entirely that higher-accuracy loop algorithms are not generally needed. This technical solution can be applied to all federated XGBoost multi-classification tasks on various data types.

Applying sparsity- and overfitting-aware eXtreme Gradient Boosting (XGBoost) for classification in federated learning allows many participants to train a series of trees collaboratively. Since various local multiclass distributions and global aggregation diversity, model performance plummets as convergence slowly and accuracy decreases. Worse still, neither the participants nor the server can detect this problem and make timely adjustments. In this article, we provide a new local-global class imbalance inconsistency quantification and utilize softmax as the activation and focal loss, a dynamically scaled cross-entropy loss, in federated XGBoost to mitigate local class imbalance. Moreover, we propose a simple but effective hyperparameter determination strategy based on local data distribution to adjust the sample weights among noncommunicating participants, synergetic focal loss, to solve the inconsistency of local and global class imbalance, a unique characteristic of federated learnin...