Abstract The rapid digital transformation of enterprises has expanded the cyber threat landscape, necessitating robust security strategies and predictive analytics to strengthen organizational resilience. This study presents an empirical analysis of enterprise cybersecurity incident data to evaluate the... The rapid digital transformation of enterprises has expanded the cyber threat landscape, necessitating robust security strategies and predictive analytics to strengthen organizational resilience. This study presents an empirical analysis of enterprise cybersecurity incident data to evaluate the effectiveness of major technical, organizational, and governance controls in reducing cyber risk and improving operational response performance. Using the Cyber Attack Detection & Risk Prediction Dataset (AI Explorer, 2026) a synthetic dataset comprising over 100,000 enterprise level incidents with 50+ features we develop a multi task machine learning pipeline employing XGBoost, Random Forest, SVM, and statistical baselines across three core tasks: binary attack success classification, multi-class risk level prioritization (Low, Medium, High, Critical), and financial loss regression. Our results reveal a hierarchical efficacy among security controls. Multi-Factor Authentication (MFA) emerges as the most impactful preventive control, reducing attack success rates by approximately 28% (37.4%  26.8%) with a medium to large effect size (Cohen’s d = 0.68). Endpoint Detection and Response (EDR) demonstrates large operational effects (d >0.8), cutting detection time by ~46 minutes and response time by ~27 minutes. Security awareness training halves successful phishing click rates (40.95%  20.91%), while weak password policies significantly elevate attack success (34.70%) and financial losses ($82,395). Firewall deployment shows modest but statistically significant benefits in reducing financial loss and cyber risk scores. Organizations aligned with ISO 27001 exhibit the lowest attack success rates (26.74%) and incident costs ($64,672). The XGBoost model achieves superior performance across all tasks: AUC-ROC of 0.961 for attack success prediction, 82.1% accuracy for risk categorization, and R² of 0.867 with MAE of $14,892 for financial loss forecasting. These findings provide a data driven blueprint for strategic cybersecurity investments, demonstrating that identity centric controls, rapid detection and response capabilities, and governance frameworks collectively form the foundation of resilient enterprise security architectures. The study confirms the practical value of integrating multiple control features and machine learning analytics for proactive cyber defense, risk quantification, and evidence based security decision making. Read More Read Less

Abstract The rapid evolution of Advanced Persistent Threats (APTs) and sophisticated evasion techniques necessitates a shift from traditional signature-based detection to intelligence-driven cybersecurity strategies. However, conventional Indicator of Compromise (IOC) approaches fail to capture the complex, underlying semantic relationships... The rapid evolution of Advanced Persistent Threats (APTs) and sophisticated evasion techniques necessitates a shift from traditional signature-based detection to intelligence-driven cybersecurity strategies. However, conventional Indicator of Compromise (IOC) approaches fail to capture the complex, underlying semantic relationships among heterogeneous threat entities. To address these limitations, this study proposes a comprehensive, graph-based Cyber Threat Intelligence (CTI) analysis framework that transforms fragmented IOCs into a structured cybersecurity knowledge graph. Utilizing the CTIDataset, comprising approximately 640,000 structured records extracted from 612 security reports, we constructed a heterogeneous graph modeling complex interactions among malware, campaigns, threat actors, and infrastructure. We evaluated multiple graph embedding techniques, including Node2Vec, Graph Autoencoders (GAE), and Variational Graph Autoencoders (VGAE), alongside advanced machine learning classifiers such as XGBoost, Random Forest, and Graph Neural Networks (GCN, GraphSAGE, GAT). Experimental results demonstrate that XGBoost achieves superior classification performance (Accuracy: 0.85, ROC-AUC: 0.91) by effectively processing complex, hash heavy tabular features and engineered attributes. Meanwhile, VGAE yields the highest embedding quality, achieving optimal cluster separation for major threat families like Burning Umbrella and Turla, as validated by Silhouette scores, Davies Bouldin indices, and comprehensive UMAP visualizations. Ultimately, this research establishes a scalable and explainable foundation for automated threat attribution and campaign discovery, bridging the critical gap between isolated indicators and semantic attack behaviors to enhance proactive cyber defense capabilities in modern digital environments. Read More Read Less

Abstract The proliferation of fake news on social media poses significant societal threats, necessitating automated detection systems that are not only accurate but also trustworthy. While recent transformer based models have achieved high classification performance, their probability calibration... The proliferation of fake news on social media poses significant societal threats, necessitating automated detection systems that are not only accurate but also trustworthy. While recent transformer based models have achieved high classification performance, their probability calibration and prediction reliability remain underexplored. This study presents a comprehensive evaluation framework for fake news detection using the large scale TruthSeeker2023 dataset, comparing traditional machine learning models (Logistic Regression, Random Forest, XGBoost) with transformer based architectures (BERT, RoBERTa). Beyond standard accuracy metrics, we systematically assess model reliability through Expected Calibration Error (ECE), Maximum Calibration Error (MCE), Brier Score, reliability diagrams, and confidence distribution analysis. Results demonstrate that RoBERTa achieves the lowest calibration error (ECE = 0.017) and superior discriminative performance (AUC  0.96), while traditional models exhibit significant miscalibration (ECE up to 0.083). Qualitative error analysis identifies persistent linguistic challenges including sarcasm (28%), missing context (24%), linguistic ambiguity (19%), and conflicting evidence (15%). Statistical significance testing (McNemar, bootstrap, DeLong, paired t-test, Wilcoxon) confirms that observed performance improvements are robust and not attributable to random variation. The findings establish that evaluating confidence reliability alongside accuracy provides a more realistic assessment of model suitability for realworld misinformation detection. This work offers evidence-based recommendations for deploying trustworthy fake news detection systems and identifies critical directions for future research, including knowledge enhanced reasoning, multimodal integration, and continual learning approaches. Read More Read Less