遗传广义加性模型

Generalized Additive Models (GAMs) achieve a balance between predictive accuracy and interpretability, yet manually configuring their structure is challenging. This research proposes the use of the multi-objective genetic algorithm NSGA-II to automatically optimize GAMs by jointly minimizing prediction error (RMSE) and a Complexity Penalty that captures sparsity, smoothness, and uncertainty. Experiments conducted on the California Housing dataset demonstrate that NSGA-II discovers GAMs that either outperform baseline LinearGAMs in accuracy or match their performance with significantly lower complexity. The resulting models are simpler, smoother, and exhibit narrower confidence intervals, thereby enhancing interpretability. This framework provides a general approach for the automated optimization of transparent, high-performance models.