Performance Comparison of Different Machine Learning Classifiers in Categorizing Bangla News Articles

In the digital age, the amount of online Bangla news has increased rapidly. For organizing their news, each news source has its own system for categorization. Because of the diversity, the structure and classification of online Bangla news items cannot always match the needs of an individual. As a result, tackling this issue and categorizing Bangla news items based on the preference of a user is a difficult task. Recommending the proper news category aids users in finding desired items, however, manually categorizing news articles is a lengthy process and cumbersome. It becomes even more challenging when addressing a language with limited resources, such as Bengali, which is one of the world’s most frequently spoken languages. However, just a few ways for categorizing Bangla news articles have been developed, with only few machine learning algorithms being used with restricted resources. Multiple machine learning approaches are highlighted in this study. Using our techniques, we were able to categorize news in nine categories with an accuracy of 93.76% in a dataset including over 400 thousand data sets from the newspaper Prothom Alo. To train diverse supervised learning models, we used dimentionality reduction and a variety of complicated textual properties, such as TF-IDF. Several machine learning classifiers were implemented. The acquired experimental findings demonstrate the recall, F1-score, precision, and accuracy of our implemented models.