A Naïve Bayes Regularized Logistic Regression Estimator for Low-dimensional Classification

What is it about?

We describe a new classifier for datasets with a small number of predictors. The method is based on logistic regression with regularization. Instead of zeroing out the coefficients of a logistic regression model, the regularization converts the coefficients to naive Bayes estimators.

Featured Image

Why is it important?

The new method performs better than logistic regression and better than naive Bayes. It can be regarded as a combination of the two methods. Extensive experiments with simulated and public-domain datasets shows that the new method is competitive with logistic regression and naive Bayes. This is especially true with datasets with a small number of predictors, where all predictors are typically used.