Machine learning algorithms for cyber threat prediction using communication authority data in emerging digital economies

Emerging digital economies face escalating cyber threats that challenge traditional security approaches, necessitating advanced predictive capabilities through machine learning technologies. This study compared machine learning algorithms for cyber threat prediction, evaluated preprocessing and feature engineering impacts, and developed an optimized model achieving precision ≥0.90 and recall ≥0.85 using Communication Authority data. Four algorithms (Random Forest, LSTM, XGBoost, SVM) were evaluated on 127,843 network traffic records spanning 18 months. Comprehensive preprocessing, feature engineering, and ensemble optimization techniques were systematically applied and validated through cross-validation and temporal analysis. The optimized XGBoost-based ensemble model achieved precision of 92.34%, recall of 89.12%, and F1 score of 90.71%, exceeding all target metrics. Preprocessing and feature engineering yielded 10.38% AUC-ROC improvement. Live deployment demonstrated 99.7% system uptime with quantified economic benefits of $3.659 million over 30 days. Machine learning approaches, particularly optimized ensemble methods combining XGBoost, Random Forest, and LSTM, provide effective cyber threat prediction for emerging digital economies, offering substantial operational and economic benefits for Communication Authority operations.