Compression-Based Averaging of Selective Naive Bayes Classifiers

摘要: The naive Bayes classifier has proved to be very effective on many real data applications. Its performance usually benefits from an accurate estimation of univariate conditional probabilities and variable selection. However, although selection is a desirable feature, it prone overfitting. In this paper, we introduce Bayesian regularization technique select the most probable subset variables compliant with assumption. We also study limits model averaging in case assumption new weighting scheme based ability models conditionally compress class labels. reduces variables, finally results "soft selection". Extensive experiments show that compression-based averaged outperforms scheme.