Performance Optimization of Naive Bayes Classification Using Filter-Based Feature Selection

Abstract

In machine learning (ML), feature selection (FS) is considered an important preprocessing step that helps to find and pick the most relevant attributes from a dataset. FS minimizes computational complexity, enhances model efficiency, and improves generalization performance by removing unnecessary features. With an important emphasis on the use of filter FS techniques in the research, this study examines the importance and effects of FS in the context of ML. FS is important in ML because it maximizes the interpretability and performance of the ML models. This study explores how well the performance of the naive Bayes (NB) classifier is influenced by filter FS methods such as Symmetrical Uncertainty, Information Gain, Gain Ratio, Chi-Square (CHISQUARE), and RELIEFF. These filter FS techniques are selected because they make ranking of features according to certain criteria and are also computationally efficient. The results captured are projected using the validity scores like accuracy, specificity, precision, false negative rate, and false positive rate using the selected features. Further, using with FS and without FS, the classifier's performance is analyzed using these validity scores. The results represent the RELIEFF approach with NBs getting the best superior results with regards to accuracy, specificity, and precision when compared to other filter FS and NB without FS strategies. Finally, the study reveals the importance of filter FS techniques in maximizing the performance and efficacy of ML models. It also provides useful information on the subtle effects of FS strategies on classification performance, offering practitioners and academics useful knowledge that will further help them optimize their models for practical uses.


Received: 5 July 2025 | Revised: 29 October 2025 | Accepted: 8 November 2025


Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.


Data Availability Statement

The data that support the findings of this study are openly available in the UCI repository at https://doi.org/10.1016/j.dss.2014.03.001, reference number [6].


Author Contribution Statement

Nalini Manogaran: Methodology, Resources. Kalpana Vadivelu: Formal analysis, Writing – original draft, Visualization. Siva Subramanian Raju: Conceptualization, Investigation, Writing – review &; editing. Yamini Bhavani Shankar: Software, Data curation, Visualization. Balamurugan Balusamy: Software, Validation, Data curation. Sumendra Yogarayan: Formal analysis, Writing – review & editing, Supervision, Project administration.