A Survey on Alleviating the Naive Bayes Conditional Independence Assumption

Naive Bayes (NB) is a simple and effective probabilistic classification method based on the Bayes theorem. NB method is used in a wide variety of real-world applications due to its effectiveness and ease of implementation. These applications include product recommendations, medical domain, credit scoring, sentiment analysis, spam filtering, etc. One of the keys to NB's efficacy is a stringent stipulation that the dataset's attributes are all treated as equal and conditionally independent. But in real-time, the NB assumption about the dataset cannot be satisfied, since the dataset captured has a high possibility of holding correlated, irrelevant, and uncertain variables. The use of these datasets does not satisfy the NB assumption, and results in poor prediction of the model. Many researchers have come up with different ways to solve the problem of the conditional independence presumption. The proposed method can be broken down into four main parts: Local Learning, Data Expansion, Feature Selection and Weighting, and Structure Extension. The goal of this study is to carry out an in-depth survey on the various methodologies that have been suggested as a means of mitigating the conditional independence presumption and determining how the performance of NB can be improved.