Short Message Service (SMS), commonly known as text messaging, is a globally adopted communication medium used by over 5 billion people. However, its simplicity and lack of built-in security have made it a prime target for phishing attacks–known as smishing–which threaten user privacy and financial security. This study presents an optimized Deep Neural Network (DNN) for detecting SMS-based phishing attempts with high reliability and efficiency. The model incorporates TF-IDF for feature extraction, SMOTE to address class imbalance, and regularization techniques including dropout and L2 penalty to enhance generalization. Evaluated on the UCI SMS Spam Collection dataset, the proposed framework achieves 99.43% accuracy and an AUC-ROC of 99.86%, outperforming more complex transformer-based and hybrid deep learning models. With only 186K parameters and a 0.18 ms inference time, the model is lightweight, fast, and well-suited for real-time mobile deployment, offering a practical and scalable solution for modern SMS phishing mitigation.