AI in Cybersecurity: Evaluating Phishing Bait Detection in Messengers

One of the most dangerous and popular attacks in digital communication is phishing. This research aims to examine how logistic regression, linear support vector machines, and decision trees perform as classic machine‑learning models on the task of detecting phishing bait in short messages on messaging platforms. A balanced corpus of phishing and legitimate short textual messages was preprocessed by tokenization, stop‑word removal, and TF-IDF (Term Frequency-Inverse Document Frequency) vectorization. The models’ performance was measured by accuracy, precision, recall, and F1-score. Our results indicate that linear SVM (Support Vector Machines) – obtains the best F1‑score 0.91, whereas the logistic regression excels in terms of accuracy 0.89, surpassing decision tree-based models by 0.04. Moreover, while accuracy is comparable to linear SVM, decision trees have significantly lower recall 0.81. Feature analysis also displays a higher response from one of the customers to terms related to urgency and, interestingly, other accounts. Consequently, our small-scale study quite effectively promoted lightweight linear classifiers are a good choice for a phishing detection high accuracy in traditional performance metrics and that an explainable artificial intelligence approach to cybersecurity is possible. Next, we’ll test larger multilingual datasets and modern deep-learning models to see whether these results hold beyond this dataset.