Predicting the risk of pipe failure using gradient boosted decision trees and weighted risk analysis

Abstract Pipe failure prediction models are essential for informing proactive management decisions. This study aims to establish a reliable model returning the probability of pipe using gradient boosted tree model, and specific segmentation grouping pipes on 1 km grid that associates localised characteristics. The is applied an extensive UK network with approximately 40,000 pipeline 14-year history. was evaluated Receiver Operator Curve Area Under (0.89), briers score (0.007) Mathews Correlation Coefficient (0.27) accuracy, indicating acceptable predictions. A weighted risk analysis used identify consequence provide graphical representation high-risk decision makers. provided important step understanding consequences predicted failure. can be directly in strategic planning, which sets long-term key decisions regarding maintenance potential replacement pipes.