City: | Saskatoon, Saskatchewan, Canada Waterloo, Ontario, Canada |
Organization: | Saskatoon Municipality Waterloo Municipality |
Project Start Date: | N/A |
Project End Date: | 20 May 2021 |
Reference: | Amini, M., & Dziedzic, R. (2021), “Comparison of Machine Learning Classifiers for Predicting Water Main Failure”, CSCE 2021 Annual Conference, Canada, 26-29 May 2021. Canada:EasyChair Preprint |
Problem: | Predicting water main pipe failure in existing water networks |
Technical Solution: | Classification models: Decision Tree, Random Forest, and Logistic Regression. |
Datasets Used: |
|
Outcome: | Comparison between the three models were conducted to select the best classification model to predict water main failure. All models gave acceptable level of accuracy for predicting pipe failure based on provided data. However, the F-1 score was low for non-homogeneous data (different pipe materials) F-1 Score for cross-validation increased after partitioning of data into homogeneous group (single pipe material) |
Issues that arose: | The use of one model for certain city didn't result in good results for other city. This indicates that data are dependent on other factors associated with the location such as temperature, soil type, soil resistivity, operational factors, ...etc. The data attributes considered in the study are not sufficient to provide a generalized model that can be used for water networks of different locations within Canada. The study outcomes are only applicable to the cities of the analysed data (i.e. Saskatoon and Waterloo). |
Status: | In Development. Further analysis is required, including other variables (such as soil type, previous rate of failure, temperature, and soil resistivity) as well as other Canadian cities. |
Entered by: | 05 Nov 2021: Rami Azzam, rami.azzam@mail.utoronto.ca. |