Aiming at the performance prediction of louvered fin heat exchangers, the predictive capabilities of various machine learning models were systematically compared. A dataset containing 343 data points was constructed through computational fluid dynamics (CFD) numerical simulations, covering three key geometric parameters sucn as fin angle, spacing and length. Heat transfer performance was further evaluated based on the j-factor, f-factor, and jf-factor. A total of 13 machine learning methods were selected, including linear models, tree-based models, ensemble models, and neural networks, with R² and RMSE used as evaluation metrics. The results showed that ensemble learning models, particularly CatBoost, performed best across all prediction tasks, with test set R² values all above 0.997 and the most stable control was RMSE, demonstrating excellent fitting capability and generalization performance. Furthermore, SHAP value analysis was conducted to reveal the influencing mechanisms of different geometric parameters and their coupling on the performance factors, providing insights for the efficient design of future heat exchangers.