Efficient model selection for kernel logistic regression

Kernel logistic regression models, like their linear counterparts, can be trained using the efficient iteratively reweighted least-squares (IRWLS) algorithm. This approach suggests an approximate leave-one-out cross-validation estimator based on an existing method for exact leave-one-out cross-validation of least-squares models. Results compiled over seven benchmark datasets are presented for kernel logistic regression with model selection procedures based on both conventional k-fold and approximate leave-one-out cross-validation criteria, demonstrating the proposed approach to be viable.