Alleviating the Naive Bayes Assumption using Filter Approaches

Particularly compared to other machine learning algorithms, the Naive Bayes algorithm is notable for its widespread application in a wide range of classification problems despite its seeming simplicity. However, the conditional assumptions it makes about the dataset are largely invalidated by the real-time data. As a result of this, the performance of the prediction is less than optimal. To solve the problem, a multitude of researchers came up with a variety of methodologies to alleviate the NB assumption. 1. Structure Learning, 2. Feature Selection, 3. Data Expansion, and 4. Attribute Weighting is the headings for the different strategies that have been proposed to mitigate the effects of the NB assumption. In comparison to the other methods, feature selection has garnered significantly more attention in recent years. Since the intention of feature selection is to find the relevant subset of variables from the complete dataset in order to improve the NB prediction and fulfill the NB presumption, it is crucial that this is accomplished without diminishing the amount of information known about the dataset. The use of feature selection in conjunction with NB is implemented in this work in order to alleviate its conditional assumption. There are several sorts of feature engineering approaches seen in nature, and employ the filter FS technique in the proposed research. Because this filter FS works quickly and effectively in comparison to other FS methodologies. In this work, two distinct Filter FS are applied, and the features that are recorded are then modeled using the NB algorithm. The experimental technique is carried out from two distinct points of view: the first perspective uses Filter FS with NB, while the second perspective uses NB only. The outcomes obtained from the two methods are evaluated, compared, and forecasted with the help of a variety of validity scores. According to the conclusion of the research, Filter-NB performs much better than NB. Also, the time complexity of filter-NB is more effective than that of NB.