Adaptive Thresholding in ML-Driven Alerting Systems for Reducing False Positives in Production Environment

Machine learning (ML)-driven alerting systems are essential for monitoring and ensuring stability in dynamic production environments. Traditional static thresholds often lead to excessive false positives, creating alert fatigue and reducing operational efficiency. This paper presents an adaptive thresholding model that dynamically adjusts alert thresholds based on real-time metrics, temporal trends, and historical data patterns. By integrating Long Short- Term Memory (LSTM) networks and autoencoders within an adaptive framework, this approach continuously learns and adapts to production data, reducing false positives and enhancing alert precision. Experimental results demonstrate that adaptive thresholding significantly reduces false alerts, improving system resilience and helping teams focus on genuine issues. 2. Keywords Adaptive Thresholding, Machine Learning, Alerting Systems, False Positives, Anomaly Detection, Temporal Analysis, Dynamic Threshold Adjustment, Contextual ML Models, LSTM, Autoencoders