ExplainableDetector: Exploring Transformer-based Language Modeling Approach for SMS Spam Detection with Explainability Analysis

High-performing AI models require explainability safeguards: The paper demonstrates that achieving 99.84% accuracy in spam detection is insufficient without transparency measures. The researchers applied Explainable AI (XAI) techniques to calculate positive and negative coefficient scores, showing that responsible AI deployment requires understanding why a model makes decisions, not just that it makes accurate ones.

Transformer-based models outperform traditional approaches for text analysis: The fine-tuned RoBERTa model significantly outperformed traditional Machine Learning models in detecting SMS spam from unstructured text data, suggesting that newer AI architectures may offer better protection against evolving cybersecurity threats when properly implemented.

Data quality and preprocessing are foundational to responsible AI: The researchers addressed class imbalance problems through text augmentation techniques and applied multiple preprocessing steps to create "clean and noise-free data." This highlights that responsible AI outcomes depend on careful data preparation, not just model sophistication.

Explainability enables model accountability in security applications: By exploring coefficient scores that show which features drive spam/non-spam classifications, the XAI analysis provides a pathway for auditing the model's decision-making process—critical for security applications where false positives or negatives have real consequences for users.