Fusion: Practice and Applications FPA 2692-4048 2770-0070 10.54216/FPA https://www.americaspg.com/journals/show/2739 2018 2018 Department of Curricula and Teaching Methods, College of Education, King Faisal University, P.O. Box: 400 Al-Ahsa, 31982, Saudi Arabia; Faculty of Specific Education, Minia university, Egypt Amr Amr Department of Mathematics and Statistics, College of Science, King Faisal University, P.O. Box: 400 Al-Ahsa, 31982, Saudi Arabia; Department of Computer Science, Faculty of Science, Minia University, P.O. Box:91519, Minia, Egypt Entesar H. Ibraheem Eliwa Department of Computer Science, Faculty of Science, Minia University, P.O. Box:91519, Minia, Egypt Ahmed Omar Online social networks continue to evolve, serving a variety of purposes, such as sharing educational content, chatting, making friends and followers, sharing news, and playing online games. However, the widespread flow of unwanted messages poses significant problems, including reducing online user interaction time, extremist views, reducing the quality of information, especially in the educational field. The use of coordinated automated accounts or robots on social networking sites is a common tactic for spreading unwanted messages, rumors, fake news, and false testimonies for mass communication or targeted users. Since users (especially in the educational field) receive many messages through social media, they often fail to recognize the content of unwanted messages, which may contain harmful links, malicious programs, fake accounts, false reports, and misleading opinions. Therefore, it is vital to regulate and classify disturbing texts to enhance the security of social media. This study focuses on building an Arabic disturbing message dataset extracted from Twitter, which consists of 14,250 tweets. Our proposed methodology includes applying new tag identification technology to collected tweets. Then, we use prevailing machine learning algorithms to build a model for classifying disturbing messages in Arabic, using effective parameter tuning methods to obtain the most suitable parameters for each algorithm. In addition, we use particle swarm optimization to identify the most relevant features to improve the classification performance. The results indicate a clear improvement in the classification performance from 0.9822 to 0.98875, with a 50% reduction in the feature set. Our study focuses on Arabic spam messages, classifying spam messages, tuning effective parameters, and selecting features as key areas of investigation. 2024 2024 08 22 10.54216/FPA.160101 https://www.americaspg.com/articleinfo/3/show/2739