An Evolutionary Amharic Fake News Detection Based on ML and Deep Learning Approach

Abstract

In recent years, the increasing dissemination of fake news threatens journalistic integrity and potentially manipulates public opinion on pivotal issues. While extensive research addresses fake news detection in dominant languages, resources for Amharic, Ethiopia’s official language, are limited. This study bridges the gap by harnessing deep learning to detect Amharic fake news. We amalgamated recent genuine and fake Amharic news articles from varied sources and combined them with the available Amharic data set to enhance the robustness of our dataset. Many machine Learning mechanisms have been employed to classify Fake and Real news. This study experimented the effectiveness of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks in classifying real and fake news specifically in the Amharic language. These recurrent neural network (RNN) architectures are well-suited for tasks like news classification due to their ability to analyze sequential data and capture long-term dependencies within text. This is particularly important for Amharic language, where word order and morphology play a crucial role in conveying meaning and identifying potential deception in fake news. LSTMs and GRUs models in Amharic fake news classification has the potential to enhance the accuracy and effectiveness of Amharic fake news detection systems. Our analysis reveals that the Gated Recurrent Unit (GRU) model achieved the highest accuracy of 98% compared to other algorithms evaluated in this study. This finding suggests that GRUs are particularly effective in the task of Amharic fake news classification. GRUs employ a gate mechanism that efficiently handles the vanishing gradient problem, a common challenge in RNNs that hinders their ability to learn long-term dependencies. This allows GRUs to effectively capture the contextual relationships between words, even when they are separated by longer distances in the text, making them particularly well-suited for Amharic language fake news classification where understanding the flow of information is crucial.