Fake news detection is a critical challenge in the digital age, where misinformation spreads rapidly, causing real-world harm. In the context of the Bangla language, this problem is exacerbated by the scarcity of labeled data for model training. This paper introduces an enriched dataset of Bangla fake news, containing 7000 authentic and 1000 fake news texts, meticulously labeled for comprehensive analysis. To tackle this issue, we explore state-of-the-art language models such as DistilBERT and RoBERTa. Our experiments with these models achieve remarkable accuracies of 94.5% and 94.3%, respectively, demonstrating the potential of transformer-based architectures for fake news detection in Bangla. Additionally, we propose a novel Deep Convolutional Neural Network, named “BFakeNewsCNN,” which, despite achieving a respectable accuracy of 85.7%, offers an efficient alternative for low-resource settings. This research contributes to the advancement of Bangla fake news detection, addressing the challenges of limited data resources while paving the way for more robust and accurate detection systems.