Predictive modelling

Decision tree model

Visual inspection

Process (computing)

Engineering

Scheduling (production processes)

Operating system

Decision process

Data mining

Computer science

Culvert

Management science

Mathematics

Mathematical analysis

Operations management

Artificial intelligence

Decision tree

Tree (set theory)

Structural engineering

Schedule

Programming language

Machine learning

Advanced Engineering Informatics

Culverts are important components of a roadway and should be properly maintained to ensure adequate road surface drainage and public safety. Culvert maintenance greatly relies on culvert inspection which is time consuming and requires a large number of skilled labor hours. Currently, State Departments of Transportation use rigid methods for scheduling culvert inspection based on one or two factors such as culvert size and/or condition. The objective of the research described in the paper is to develop a more intelligent scheduling system for culvert inspection to improve the utilization of limited resources. The proposed intelligent system first predicts the conditions of the culverts that are due for inspection in a given year and based on the prediction results, only schedule inspections for those predicted to be in poor condition. The prediction models utilized a Decision Tree algorithm together with the Synthetic Minority Over-Sampling Technique to deal with the highly imbalanced data in the culvert inventory database. The case study presented in the paper utilized 12,400 culvert records from the Ohio Department of Transportation to train and test the prediction models. The developed prediction models have achieved accuracies over 80% for the training set and 75% for the testing set and satisfactory values for the areas under the curve of 0.8. The case study concluded that by implementing the proposed intelligent culvert inspection scheduling system, the number of culverts needing inspections is reduced by 44%. Implementation of the proposed system could assist state and local agencies with prioritizing inspection of culverts needing attention while maximizing the use of limited resources. While this study is applied to culverts in Ohio, the proposed framework can be used on any similarly available culvert data set worldwide. The paper ends by providing suggestions to improve the quality of the data in culvert inventory databases.

The use of decision tree based predictive models for improving the culvert inspection process