Abstract

This study develops a hybrid AI model integrating LSTM, Random Forest, and XGBoost with 
Bayesian inference and Z-score analysis to enhance cybersecurity threat prediction in financial 
transactions. The model achieves 100% accuracy, precision, recall, and F1-score (Tables 3 and 4), 
outperforming standalone machine learning models. Bayesian inference estimates a fraud probability 
of 0.3213 (Table 5), while Z-score analysis detects zero anomalous transactions (Table 6, Figure 2), 
confirming its reliability. The ROC curve (Figure 3) validates the model’s strong discriminatory 
power, reducing false positives and negatives. These findings highlight the effectiveness of hybrid AI 
models in fraud detection, ensuring both accuracy and interpretability. This study’s implication is that 
financial institutions can significantly reduce cyber fraud risks by adopting AI-driven fraud detection 
systems. It is recommended that financial sectors integrate risk-based Bayesian fraud probability 
scoring to improve transaction monitoring and enhance cybersecurity resilience.