Author(s): Luka Vinokic; Milos Milasinovic; Zeljko Vasilic; Veljko Prodanovic; Milan Gocic
Linked Author(s):
Keywords: Anomaly detection; Data quality; Digital twin; Sensors; Urban drainage system
Abstract: This study introduces a novel agreement-based algorithm designed using two distinct machine learning (ML) approaches for anomaly detection and data recovery to improve the reliability of sensor data streams, particularly within complex monitoring environments such as urban drainage systems. The proposed algorithm integrates ensemble-based confidence analysis, anomaly classification, and a dynamic agreement mechanism that iteratively reconciles conflicting detector outputs. By combining two distinct outlier-detection principles utilizing ML decision-making, the approach enhances robustness in automated data quality control.
Year: 2026