Video game addiction is an emerging global concern with significant impacts on individual well-being. This study develops a machine learning-based framework to detect potential addiction traits using data from Reddit, where 987 user posts were analyzed. The goal was to characterize signs of video game addiction and develop predictive models. Various text representation and word embedding techniques, including Bag of Words (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), Word2vec, and BERT, were employed alongside analyses of emotional expression and topical content in the posts. These posts were preliminarily labeled by psychologists to facilitate the training of models. Four supervised classification algorithms—Logistic Regression, KNN, Decision Tree, and AdaBoost—were utilized for model evaluation. The study highlights the efficacy of embedding-based models, particularly the combination of Word2vec and Logistic Regression, which achieved the highest accuracy of 0.94. These findings advance our understanding of how machine learning can be leveraged to identify behavioral patterns associated with video game addiction in social media contexts, pointing towards potential applications in clinical and psychological diagnostics.