An Efficient Feature Selection Strategy for Enriching the Precision of Machine Learning Models

Machine learning (ML) based training and evaluation for high-dimensional text processing requires extensive computational resources and longer execution times. Effectively managing feature relevance is vital for reducing computational demands and is essential for progressing natural language processing research and realizing its full potential, especially in Bengali text processing. This work proposes a robust strategy utilizing four feature selection classifiers to determine crucial features and evaluate their effectiveness in reducing computational bottlenecks. The proposed strategy highlights that it consistently outperforms the underneath classifiers across different feature sets, achieving higher accuracy with increasing features. The proposed approach emphasizes the most elevated fifty attributes with the best metrics: 90% accuracy, 89.09% precision, and 92.45% recall. The presented strategy exhibits its significance through a detailed comparison with related work, demonstrating a significant improvement of 4.35%.