Towards Dimension Reduction: A Balanced Relative Discrimination Feature Ranking Technique for Efficient Text Classification (BRDC)

The volume and complexity of textual data have significantly increased worldwide, demanding a comprehensive understanding of machine learning techniques for accurate text classification in various applications. In recent years, there has been significant growth in natural language processing (NLP) and neural networks (NNs). Deep learning (DL) models have outperformed classical machine learning approaches in text classification tasks, such as sentiment analysis, news categorization, question answering, and natural language inference. Dimension reduction is crucial for refining the classifier performance and decreasing the computational cost of text classification. Existing methodologies, such as the Improved Relative Discrimination Criterion (IRDC) and the Relative Discrimination Criterion (RDC), exhibit deficiencies in proper normalization and are not well-balanced regarding distinct class's term ranking. This study introduced an improved feature-ranking metric called the Balanced Relative Discrimination Criterion (BRDC). This study measured document frequencies into term-count estimations, facilitating a normalized and balanced classification approach. The proposed methodology demonstrated superior performance compared to existing techniques. Experiments were conducted to evaluate the efficacy of the proposed techniques using Decision Tree (DT), Logistic Regression (LR), Multinomial Naïve Bayes (MNB), and Long Short-Term Memory (LSTM) models on three benchmark datasets: Reuters-21578, 20newsgroup, and AG News. The findings indicate that LSTM outperformed the other models and can be applied in conjunction with the proposed BRDC approach.

@article{Nasir2024,
  title     = {Towards Dimension Reduction: A Balanced Relative Discrimination Feature Ranking Technique for Efficient Text Classification (BRDC)},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {7},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Muhammad Nasir and Noor Azah Samsudin and Wareesa Sharif and Souad Baowidan and Dr. Humaira Arshad and Muhammad Faheem Mushtaq},
  doi       = {10.14569/IJACSA.2024.0150766},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150766}
}