Robust Anomaly Recognition in Hydraulic Structural Safety Monitoring: A Methodology Based on Deconfounding Boosted Regression Trees

Table 1

Regression modeling approaches consideration on adjacent response points and their limitations.

Modeling approach	Limitations
Without consideration	Insufficient monitoring of environmental factors, the regression model is prone to missing alarms
Direct modeling	The regression model suffers from severe overfitting and is prone to false alarms
Two-stage regression	The powerful nonlinear fitting capability of machine learning hinders overfitting correction
Separate regression	The integrated regression model ignores interactions between causal factors and correlation factors