Combining XGBoost with Particle Swarm Optimization to Improve Phishing Detection

An improved phishing detection system, an important tool in cyber security, can help reduce the occurrence of phishing cyber-crimes. Existing phishing detection tools such as machine learning models have shown its capability in detecting phishing attempts, but still face some limitations in detection accuracy due to suboptimal hyperparameters. One of the existing powerful high performance and scalability classification model, such as XGBoost shows great result in phishing detection. By optimizing the hyperparameters using PSO, this paper exploits it to further increase the detection accuracy of the XGBoost model. The result of the proposed method XGBoost-PSO model is compared with the original XGBoost model created based on previous research. The proposed method shows that it is able to improve the detection accuracy from 97.53% to 97.67%, recall increases from 93.31% to 98.84%, and the F1-score improves from 97.8% to 97.93%, making the XGBoost-PSO model more effective at minimizing false negatives and false positives. Although precision slightly decreases from 97.3% to 97.03%, the overall improvement in the F1-score demonstrates the effectiveness of hyperparameter optimization in phishing detection.