The detection of smishing messages (SMS-based phishing) compared to legitimate SMS content remains a critical challenge in cybersecurity. This study explores the integration of textual embeddings and stylometric features to improve the classification of smishing and legitimate messages. Using a diverse dataset of over 22000 SMS messages, we applied preprocessing steps, including tokenization and lemmatization, extracting stylometric and presence-based features, and creating embeddings such as Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec. Experiments were conducted with classifiers including Logistic Regression, Random Forest, and XGBoost. Results demonstrated that using both Word2Vec embeddings and stylometric features provided the best cross-validation accuracy of 97.10% with the XGBoost classifier. TF-IDF embeddings paired with stylometric features also achieved cross-validation accuracy improvements, reaching 96.24% with XGBoost. Stylometric features alone provided moderate performance, emphasizing their complementary role when merged with embeddings. This research underscores the importance of feature engineering in capturing both linguistic patterns and structural characteristics for detecting smishing messages. The findings highlight the potential of hybrid approaches for robust phishing detection across varied contexts.