A Comparative Study of Machine Learning Applications in Enterprise Network Fault Detection and Self-Healing Infrastructure (2018–2026)

Abstract

This study conducted a comprehensive quantitative comparative analysis of machine learning applications in enterprise network fault detection and self-healing infrastructure over the period 2018 to 2026. The research adopted a retrospective cross-sectional design, analyzing 124 empirical studies sourced from major academic databases to evaluate the performance and applicability of different machine learning approaches. The analysis focused on key model categories, including traditional machine learning, deep learning, hybrid and ensemble models, unsupervised techniques, and reinforcement learning, across multiple enterprise network domains such as telecommunications, cloud computing, software-defined networks, and IoT-based systems. The findings revealed that deep learning models achieved the highest overall detection performance, with an average accuracy of 94.2%, followed by hybrid models at 92.6%, while traditional machine learning models recorded comparatively lower accuracy at 88.4%. Unsupervised methods demonstrated moderate effectiveness, particularly in anomaly detection scenarios, whereas reinforcement learning showed distinct advantages in recovery-related metrics, achieving the lowest mean time to repair of 19.6 seconds and the highest system uptime of 98.3%. Domain-level analysis indicated that cloud computing and software-defined networks outperformed other environments, achieving accuracy levels of 94.8% and 93.6%, respectively, while IoT and edge systems lagged behind with an average accuracy of 88.5%. Subgroup analysis further demonstrated that studies using benchmark datasets reported higher accuracy levels of 93.8% compared to 89.6% for real-world datasets, highlighting the influence of data realism on model performance. Additionally, k-fold cross-validation yielded more robust results, with an average accuracy of 92.9%, compared to 89.7% for holdout validation. Inferential statistical analysis confirmed that these differences were statistically significant, with effect sizes ranging from moderate to large across key performance indicators. Overall, the study concluded that advanced machine learning architectures, particularly deep learning and hybrid models, combined with reinforcement learning for recovery optimization, offer the most effective solutions for enterprise network fault detection and self-healing systems. The findings provide both theoretical insights and practical guidance for designing scalable, data-driven, and resilient network management frameworks.