Computer Science · Information Systems
Classification of Phishing Attacks from OSINT Data Using Naive Bayes Algorithm for Enhanced Cyber Situational Awareness
Sahat Erwin Gemayel Siagian, Herry Sujaini, Rudy Gianto
See on the mapAbstract
Phishing attacks are among the most frequent and rapidly evolving cyber threats, often serving as the primary entry point for large-scale security incidents. Efficient and accurate phishing detection is essential to support Cyber Situational Awareness (CSA), enabling rapid response and early mitigation of threats. This study proposes a phishing detection model based on Open Source Intelligence (OSINT) using the Naive Bayes algorithm, chosen for its lightweight and efficient nature. The dataset used is the UCI Phishing Websites Dataset, consisting of 30 features and 11,055 entries, analyzed and validated through an OSINT perspective. Evaluation was performed using 10-fold cross-validation with accuracy, precision, recall, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-score, and ROC-AUC metrics. The results show that the model achieved an accuracy of 92.98 %, precision of 0.936, recall of 0.904, F1score of 0.919, and ROC-AUC of 0.981, with a training time of only 0.03 seconds. These findings indicate that the proposed model is reliable for implementation in environments requiring fast detection with limited resources. However, this study also acknowledges its limitations, particularly the presence of false negatives and reliance on a static dataset. Therefore, future research will focus on real-time testing within CSA environments, expansion to multilingual datasets, and comparisons with modern deep learning models to further validate the advantages of the proposed approach.