A Novel Feature Selection Technique for Text Classification Using Naïve Bayes.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Text classification is crucial for analyzing unstructured data, with applications in sentiment analysis and spam detection.
• Naïve Bayes (NB) is a simple, popular text classification algorithm but often underperforms compared to other methods.
• Existing limitations make NB unusable despite its intuitive nature and low data requirements.

Purpose of the Study:

• To propose a novel two-step feature selection method to improve Naïve Bayes performance in text classification.
• To address the poor performance of Naïve Bayes by enhancing its feature selection process.
• To make Naïve Bayes a competitive classifier through improved feature selection.

Main Methods:

• A two-step feature selection process combining univariate feature selection and feature clustering.
• Univariate feature selection is used to reduce the initial feature search space.
• Feature clustering is applied to select a set of relatively independent features.

Main Results:

• The proposed method significantly improves Naïve Bayes performance across 13 diverse datasets.
• Evaluations show the enhanced Naïve Bayes classifier achieves performance comparable or superior to other leading classifiers.
• The algorithm outperforms traditional feature selection methods like greedy search and CFS.

Conclusions:

• The proposed two-step feature selection method effectively enhances Naïve Bayes for text classification tasks.
• This approach revitalizes Naïve Bayes, making it a viable and competitive option for text categorization.
• The method offers a practical solution to improve the utility of a simple yet intuitive classification model.