The rapid spread of rumors due to the growth of the internet and social media has prompted researchers to search for solutions to detect fake news, which is treated as a text classification problem. In this study, we examine the content of fake news in the Arabic-speaking world through YouTube comments. To start, we have updated our Arabic corpus for fake news analysis, incorporating the most frequently discussed topics in rumors. Next, we conducted experiments to determine the best combination of preprocessing, classical machine learning, and neural networks in classifying comments as either rumors or non-rumors. The models we used include Support Vector Machine (SVM), Decision Tree (DT), Multinomial Naïve Bayes (MNB) for machine learning, and LSTM, BILSTM, and CNN for deep learning. Finally, we compare the results of previous machine learning models and deep learning techniques to determine which one is more effective in detecting fake news. Both models showed high accuracy in the results.