Building an Ensemble of Fine-Tuned Naive Bayesian Classifiers for Text Classification

Area of Science:

• Machine Learning
• Natural Language Processing
• Computational Linguistics

Background:

• Naive Bayesian (NB) algorithms excel in text classification.
• Ensemble methods typically offer limited gains with stable NB classifiers.
• Improving NB classifier performance via ensembles is challenging.

Purpose of the Study:

• To investigate ensemble-building techniques for enhancing NB text classification.
• To develop a fine-tuning algorithm for NB classifiers that improves accuracy and stability.
• To evaluate the suitability of the proposed algorithm for ensemble methods like bagging.

Main Methods:

• Proposed a novel fine-tuning algorithm for NB classifiers (FTNB).
• Compared FTNB ensembles with standard NB ensembles.
• Utilized bagging for ensemble construction.
• Conducted empirical experiments on 16 benchmark text classification datasets.

Main Results:

• Ensembles of standard NB classifiers showed minimal accuracy improvement.
• Ensembles of fine-tuned NB (FTNB) classifiers achieved significant accuracy gains.
• The FTNB algorithm demonstrated improved accuracy and reduced stability compared to standard NB.
• FTNB proved more suitable for ensemble methods like bagging.

Conclusions:

• Fine-tuning NB classifiers is a viable strategy for improving text classification accuracy.
• Ensembles of fine-tuned NB classifiers, especially with bagging, offer substantial performance benefits.
• The proposed FTNB algorithm enhances NB suitability for ensemble learning in text classification.