Efficient Online classification using an Ensemble of Bayesian Linear Logistic Regressors

Abstract

We present a novel ensemble of logistic linear regressors that combines the robustness of online Bayesian learning with the flexibility of ensembles. The ensemble of classifiers are built on top of a Randomly Varying Coefficient model designed for online regression with the fusion of classifiers done at the level of regression before converting it into a class label using a logistic link function. The locally weighted logistic regressor is compared against the state-of-the-art methods to reveal its excellent generalization performance with low time and space complexities.