SMS Spam Detection with NLP and Ensemble Learning for Enhanced Classification

The rapid rise in Short Message Service (SMS) communication has also seen a surge in spam content, including phishing and deceptive advertisements. This study proposes a robust and scalable spam detection framework that integrates advanced Natural Language Processing (NLP) techniques with an ensemble that fuses classical machine learning models and deep learning models to enhance classification precision over baseline methods [3] [5] [11] [12] [13]. The system incorpo-rates comprehensive preprocessing—tokenization, stemming, and lemmatization—paired with a hybrid feature extraction approach using Term Frequency-Inverse Document Frequency (TF-IDF) and CountVectorizer. We evaluated multiple classifiers, notably Random Forest, Support Vector Machine (SVM), and Extra Trees, alongside deep learning models (Bidirectional Long Short- Term Memory, Gated Recurrent Unit, and a transformer-based model) in a soft-voting ensemble, achieving a peak accuracy of 98.92% and an F1-score of 0.987 [9] [4]. To ensure model transparency, Explainable AI tools such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were integrated, offering interpretability in real-time predictions [1] [2]. The solution is deployed via a Flask API with a simple web-based interface, demonstrating near-real-time SMS classification, and is extensible to full real-time throughput with Docker and message queues. This work contributes a concrete, interpretable, and industry-ready tool for securing communication platforms—especially in banking, e-commerce, and telecommunications—where spam incurs sig-nificant operational risks. Future work will explore multilingual datasets, adversarial robustness, and continuous learning to further strengthen system adaptability.