Abstract
This quantitative investigation addresses the application of artificial intelligence (AI) models for predicting cyberattacks and detecting anomalies in network traffic, aiming to strengthen cybersecurity defenses. As cyber threats grow in complexity, AI provides significant opportunities for predictive and responsive protection. This study compares three AI models ‒ Ensemble Methods, Isolation Forest, and ARIMA ‒ using datasets aggregated on daily, weekly, and monthly levels. The methodology covers advanced data preprocessing, statistical analysis, and evaluation metrics such as RMSE, R², Precision, Recall, and F1-Score. Ensemble Methods demonstrated outstanding accuracy and reliability, achieving high R² values and minimal errors. Isolation Forest was effective in identifying anomalies and detecting outliers, despite its limitations in explaining broader data variability. ARIMA showed potential in time-series analysis but required optimization to improve precision and reduce false positives. These findings emphasize the importance of combining ensemble techniques with other approaches to improve the accuracy and adaptability of AI models in dynamic cybersecurity environments.