The rapid advancement of information technology has reshaped human interaction, embedding platforms like Facebook, X (formerly Twitter), Instagram, and WhatsApp into daily life. While these platforms enhance connectivity, they also facilitate the spread of hate speech, threatening social harmony. This study compares the performance of two text classification models—Support Vector Machine (SVM) with four kernels and Convolutional Neural Network (CNN) in detecting Indonesian hate speech on platform X. A dataset of 11,488 entries was collected through manual crawling and processed using TF- IDF and Word2Vec. Experiments were conducted across four scenarios: varying train-test ratios, N-Gram types, N-Gram combinations, and Word2Vec vector dimensions. Results show that CNN with TF-IDF and Unigram + Bigram achieved the best accuracy of 82.90%, while SVM with a Sigmoid kernel reached 81.94%. These findings support the development of automated and reliable hate speech detection systems for social media.