Culvert Condition Prediction via Artificial Neural Network Machine Learning-Based Models using SMOTE

Underground stormwater infrastructure, such as culverts, present a significant maintenance challenge for municipal agencies due to aging, urbanization, and economic pressures. Decision support via machine learning regarding the most effective renewal, replacement, and maintenance, dependable condition prediction can alleviate this burden. In contrast to conventional mathematical models, machine learning-based models generally offer better performance when processing large datasets with missing or “noisy” data. A novel data-driven approach for predicting culvert conditions based on existing data inventory is proposed in this paper using an artificial neural network with a synthetic minority oversampling technique to address imbalanced datasets. Preliminary results show the viability of the proposed machine learning framework for use with this work’s application.