AI-driven predictive analytics for response time estimation in fraud detection systems: A production-scale decision support study

Fraud detection systems operate under strict latency requirements while processing large and dynamically varying transaction volumes. In such mission-critical environments, response time directly influences transaction success, customer experience and regulatory compliance. Traditional performance monitoring approaches are predominantly reactive and provide limited capability for anticipating performance degradation. This paper presents an AI-driven predictive analytics framework for estimating response time in a production-scale fraud detection system using real operational data. The proposed approach formulates response time estimation as a supervised regression problem based on system utilization metrics, workload intensity and error characteristics collected from a live production environment. Multiple machine learning models are evaluated to capture both linear and non-linear performance behavior. Experimental results demonstrate that ensemble-based models significantly outperform baseline approaches, highlighting the effectiveness of data-driven techniques for performance prediction. The framework further integrates predictive insights into a decision-support context, enabling proactive performance management, capacity planning and SLA risk mitigation. The study demonstrates the practical value of AI-driven predictive analytics for enhancing performance assurance in real-world fraud detection systems.