Optimizing Email Spam Detection in Academic Environments: A Machine Learning Approach

In the digital age, spam detection remains a critical challenge, particularly in educational environments where the nature and volume of communication differ significantly from other domains. The increasing frequency and sophistication of email-based hacking attacks pose significant security threats to academic institutions. This paper presents an approach to detecting spam in an educational context using machine learning algorithms. By leveraging a dataset collected from an academic institution, we highlight the unique characteristics of academic communication and their implications for spam detection. Our methodology involves a comprehensive feature selection technique to identify the most relevant attributes for effective spam filtering. To address the imbalance in the dataset, class balancing techniques are employed, ensuring the machine learning models are trained on a representative distribution of spam and non-spam messages. The impact of feature selection on the classification of spam emails is investigated using five machine learning algorithms. Evaluation of the classifiers determines the most effective approach for classification of spam emails in the educational context. The experimental results demonstrate that our approach achieves remarkable accuracy, with the Random Forest classifier reaching up to $\mathbf{9 9. 9 \%}$ accuracy. This high level of precision underscores the potential of tailored machine learning solutions in educational spam detection and sets a benchmark for future research in this area.