Cyber Threats Prediction And Analysis Using Machine Learning Algorithms

In the ever-shifting terrain of cybersecurity, threat prediction and analysis became central to creating a lateral defense mechanism. The work presents a strong comparison of five machine learning algorithms, viz. Random Forest, Support Vector Machine (SVM), Naive Bayes, XGBoost, and Decision Trees for cyber-threat prediction and classification. The past few years have seen the development and maintenance of a large-scale dataset containing various cyber threat indicators and network behavior patterns. After several iterations and optimizations, the Random Forest seemed to do best with the threat predictive classification of 80.88% accuracy and an F1-score rating of 0.84. The model performed extraordinarily well in classifying high-severity threats because, for some threat categories, the precision increases went up to 95%. On top of that, hyperparameter tuning on the Random Forest model was applied in the study, increasing prediction accuracy and reducing false positives even further. The comparative study also offered some insights regarding training time for Random Forest, XGBoost, and SVM models with Naive Bayes being the fastest at 9.8 seconds with decent accuracy. The study contributes to the incidents by establishing a systematic approach to cyber threat prediction and by giving some hints for practical applications for machine learning-based security solutions.