A Magnetic Core Loss Prediction Method Based on Multi-Feature Dimensionality Reduction and XGBoost Five-Fold Crossover Model

High-frequency magnetic components are widely used in various equipment in power electronics. The loss characteristics of magnetic core components are key parameters that significantly affect their performance, with higher losses leading to lower converter efficiency. To improve the accuracy of core loss prediction, we proposed a magnetic core loss prediction method based on multi-feature dimensionality reduction and an XGBoost five-fold cross-validation model. Key features such as material, frequency, temperature, waveform type, peak magnetic flux density, and effective magnetic flux density are analyzed and identified. Multi-feature dimensionality reduction is performed using Pearson correlation analysis, principal component analysis, and SelectKBest, resulting in three selected input features. A five-fold cross-validation prediction model based on XGBoost is constructed, and core loss prediction research is conducted using the obtained sample data. The results show that the proposed method achieves a mean absolute percentage error (MAPE) of 120.2407, significantly outperforming other comparative models in prediction performance, providing a valuable reference for improving magnetic component loss prediction methods.