Attribute and instance weighted naive Bayes

Naive Bayes (NB) continues to be one of the top 10 data mining algorithms, but its conditional independence assumption rarely holds true in real-world applications. Therefore, many different categories of improved approaches, including attribute weighting and instance weighting, have been proposed to alleviate this assumption. However, few of these approaches simultaneously pay attention to attribute weighting and instance weighting. In this study, we propose a new improved model called attribute and instance weighted naive Bayes (AIWNB), which combines attribute weighting with instance weighting into one uniform framework. In AIWNB, the attribute weights are incorporated into the naive Bayesian classification formula, and then the prior and conditional probabilities are estimated using instance weighted training data. To learn instance weights, we single out an eager approach and a lazy approach, and thus two different versions are created, which we denote as AIWNBE and AIWNBL, respectively. Extensive experimental results show that both AIWNBE and AIWNBL significantly outperform NB and all the other existing state-of-the-art competitors. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

Naive Bayes algorithm remains one of the top 10 data mining algorithms, but its conditional independence assumption is rarely true in real-world applications. This study introduces a new improved model called AIWNB, which combines attribute weighting and instance weighting into one uniform framework, to address this issue effectively.