MACHINE LEARNING-BASED PROCESS MINING FOR ANOMALY DETECTION AND QUALITY ASSURANCE IN HIGH-THROUGHPUT MANUFACTURING ENVIRONMENTS

Abstract

This study investigated the effectiveness of a machine learning–based process-mining framework for anomaly detection and quality assurance in high-throughput manufacturing environments. The research aimed to determine whether integrating process-mining metrics with data-driven machine learning techniques could outperform the traditional Statistical Process Control (SPC) system in predicting product non-conformance and operational inefficiencies. A quantitative research design was employed using a dataset of 142,368 production instances collected from Manufacturing Execution System (MES), Supervisory Control and Data Acquisition (SCADA), and Quality Management System (QMS) logs across three automated production lines. Variables included cycle-time variation, waiting time, machine utilization, rework frequency, conformance fitness score, and sensor-derived telemetry aggregates. A binary logistic regression model was developed to estimate the probability of product non-conformance. The model achieved a statistically significant fit (χ² = 482.76, p < 0.001) and explained 46.2% of the variance in defect occurrence (Nagelkerke R² = 0.462). Key predictors included Conformance Fitness Score (β = –0.72, p < 0.001) and Rework Frequency (β = 0.58, p < 0.01), indicating that lower process conformance and higher rework activity substantially increased the likelihood of defects. Comparative performance analysis showed that the machine learning–based model achieved higher predictive accuracy (91.3%) than the SPC baseline (84.9%), along with improved AUROC (0.93 vs. 0.82) and AUPRC (0.48 vs. 0.41) scores. These findings demonstrated that the proposed model provided superior anomaly detection capability, reduced false-alarm rates, and enhanced predictive precision. Overall, the integration of machine learning and process-mining analytics significantly improved operational reliability and quality assurance performance. The study concluded that adopting such intelligent process-monitoring systems can strengthen defect prevention, streamline production decision-making, and support real-time process optimization in smart manufacturing environments. The results contributed to advancing data-driven quality engineering and reinforced the role of AI-powered analytics in modern industrial process control.