Explainable AI-Driven Bangla News Classification: Comparative Study of ML and DL Approaches

Classifying Bangla news is still a big problem in processing languages with few resources, however most recent studies emphasize performance metrics instead of model explainability. This study introduces an explainable AI-based system for the multiclass classification of Bangla news articles into four categories: Sports, International, Entertainment, and National, utilizing a dataset of 11,904 articles. We conducted a comprehensive comparison of classical machine learning (ML) and deep learning (DL) methodologies. We used Ridge Classifier, SGD Classifier, Multinomial Naive Bayes, and XGBoost to help us learn how to use machines. We got the best score of 92.55% accuracy by using a hard-voting group of the four best linear models. We make deep learning models like CNN-LSTM, Simple LSTM, Stacked LSTM, and GRU. When they work together, the three best models get 95.76% right, and when they work alone, they get 95.46% right. We used LIME (Local Interpretable Model-agnostic Explanations) to make the Ridge Classifier and Multinomial Naive Bayes models more accurate and easier to understand. The results show that DL ensembles are better at finding the right answer than more common ML models. However, linear ML models are faster and easier to learn, train, and use. This change made Bangla NLP systems more open and reliable. This is a great way to stay up to date on the news and learn about what's going on in places where Bengali is spoken.