The rapid advancement of technology made the transition of information too simplified. Especially, the growth in the number of social media platform users has accelerated this trend. News and trends on social media nowadays control mainstream news platforms or even the media which makes it important to distinguish fake news from the real one. In the regions where Bangla is mostly spoken, mainly in Bangladesh, circulation of misinformation is so common, and it sometimes raises conflict on sensitive subjects such as religion or racism. As anyone can share knowledge online, and the internet is everywhere in the country, people share whatever information they want to share regardless of the authenticity. In the rural areas where the literacy rate is low, people seem to consume this unauthentic information. Which amplifies the need of automatic detection of fake news. In recent years, with the advancement of machine learning models, many researchers have found ways to classify fake news, most of which works for news in the English language. Even though some work has been done for detecting fake news in Bangla, it is just not enough. Computational resource is not widely available in the countries like Bangladesh which makes it important to find a proper Machine Learning (ML) model to classify fake news accurately with optimal resource utilization. Machine learning algorithms are popular for their fast training time and less resource utilization. This paper shows a comparison of the performance, training time and memory usage of six different traditional machine learning algorithm-based models and suggests SVM as the best performing with low resource utilization.