Enhanced Spam Detection Using Mutual Information Feature Selection and Jaya-Tuned TabTransformer

Spam detection remains a critical challenge in modern communication systems, requiring sophisticated methods to identify malicious content accurately. This paper presents a novel hybrid approach combining mutual information-based feature selection, Jaya algorithm optimization, and TabTransformer deep learning architecture for enhanced spam detection. Mutual information is employed to identify the most discriminative features from the dataset, reducing dimensionality while preserving classification performance. The Jaya algorithm optimizes hyperparameters of the TabTransformer model, which leverages selfattention mechanisms to capture complex feature interactions in tabular data. Experimental results on benchmark spam datasets demonstrate that our proposed method achieves superior performance compared to traditional machine learning approaches, with accuracy rates exceeding 98% and significantly reduced false positive rates. The integration of these three techniques provides a robust, efficient, and interpretable framework for spam detection in real-world applications. Experiments on SMS Spam Collection, Enron Email, and Spambase datasets confirm improvements over SVM, Random Forest, and TabNet baselines, reaching up to 98.3% accuracy while lowering false positives.