Many individuals are currently migrating away from their current place as a result of fake news spread on social media platforms such as Facebook, Telegram, Imo, and WhatsApp. Due to this, the Ethiopian government and people face several problems, including food, water, electricity, and transportation. As a result, in this paper, we provide a deep learning-based Amharic fake new detection model that can distinguish between the genuine and the fake. In addition, we used FastText pre-trained word embedding to generate vector representations of all the features used in the dataset and to preserve the sematic and syntactic relationships found among words. The dataset for this study was obtained from the Ethiopian Prime Minister's official Facebook page, while the fake news was obtained from the Army Force website. According to the testing results, the proposed convolutional neural network (CNN) based false news detection algorithm produced an accuracy of 98.82%. Therefore, the model and datasets presented here might assist governments, decision-makers, and public judges in determining the reliability of information broadcast on social media during the conflict.

Keywords: deep learning; FastText; fake news; text classification