With the great expansion of technology, there has become a need for a tool that detects aggressive behavior on social media sites to prevent crimes from occurring in the future. This research focused on analyzing the general data processing model for social media platforms and for enhancing performance by applying natural language processing, feature extraction techniques, and machine learning algorithms. Text data was gathered from Facebook and Instagram, comprising texts in both Arabic and English. The Arabic texts were translated into English, resulting in a dataset of 150 English rows. Natural language processing techniques were employed for data preprocessing, while feature extraction utilized methods like term frequency-inverse document frequency, N-grams, and Bag of words. The research implemented various machine learning algorithms including Naive Bayes and Bag of Words with Naïve Bayes achieved notable accuracy. Furthermore, the performance of the enhanced model was compared with other experiments conducted on two external datasets, demonstrating superior effectiveness.