ORCID Identifier(s)


Graduation Semester and Year




Document Type


Degree Name

Doctor of Philosophy in Civil Engineering


Civil Engineering

First Advisor

Mohammad Najafi


Wastewater collection systems deteriorate over time, requiring continuous adjustments and the development of asset management frameworks on the part of utility owners to maintain the performance of their assets. Any asset management framework should emphasize the importance of asset inspection and condition evaluation for efficient system operation and maintenance. Closed-circuit television (CCTV) is the most widely used tool in the United States for inspecting the interior of sewer pipes, which is a somewhat expensive and time-consuming process given the extensive inventory of pipes in a city. Due to their vast inventory of these pipes, no municipality can inspect every individual sanitary sewer pipe section in a short amount of time. Therefore, the main goal of this research is to develop prediction models that can anticipate the future state of sewer pipes. The results of the models can be used to rank the necessity for sanitary sewer pipe inspection, rehabilitation, and replacement. Combined data collected from the City of Dallas, Texas, and the City of Tampa, Florida, were used in this dissertation. This dataset included nine independent variables: pipe age, size, length, material, surrounded soil type, soil pH, depth, slope, and surface conditions, and one dependent variable was the condition rating of sewer pipe based on PACP scores from 1 to 5. Different resampling procedures were examined in this study to overcome the problem of the imbalanced dataset, and finally, the resampled dataset by the SVM-SMOTE method was selected. Various machine learning algorithms such as Logistic Regressions, k-nearest neighbors, Decision trees, Random Forests, AdaBoost, Gradient Boosting Tree, and XGBoost were employed to develop prediction models. The other objective of this dissertation is an investigation of the efficiency of different machine learning methods using a resampled dataset, which was done thoroughly in this study. Various evaluation metrics, including precision, recall, F1-score (see Section 3.5.5), and area under the curve (AUC), were calculated to compare the effectiveness of developed models. The overall F1-score for the Random Forest model was 0.80 and for Multinomial Logistic Regression was 0.48, which were the highest and lowest, respectively. It was concluded that tree-based models had better performance than other models and the bagging approach was more efficient than boosting. Additionally, as another objective of this dissertation, using the best model results, it was found that pipe age and length had the highest effect on the condition rate of sewer pipes, while pipe location had the least impact.


Sewer pipe, Asset management, Prediction model, Machine learning


Civil and Environmental Engineering | Civil Engineering | Engineering


Degree granted by The University of Texas at Arlington