Abstract:

The pervasive use of email as a critical communication medium has made it a prime target for cyberattacks, particularly phishing. Traditional filtering systems often fail to distinguish sophisticated phishing attempts from legitimate correspondence. This paper presents a hybrid machine learning model that synergistically combines a Naïve Bayes classifier with a Support Vector Machine (SVM) to enhance phishing detection capabilities. The model analyzes a comprehensive feature set extracted from email headers, body content, and embedded URLs. Trained and tested on a dataset of 5,229 emails from EthioTelecom and public phishing repositories, the proposed system demonstrates high efficacy. Experimental results indicate that the SVM component slightly outperforms Naïve Bayes, achieving an overall accuracy of 98.57%, a precision of 97.8%, a recall of 98.83%, and an F1-score of 97.83%. The study concludes that hybrid models offer a robust and adaptive solution to the evolving threat of phishing, addressing significant gaps in current email security frameworks.