Downloading Data from LTPP Website to use in Stochastic Model
Author: Michael Rechtin Major: Industrial Engineering
During the Fall semester of 2020, I continued my research in the Industrial Engineering Department under my advisor, Dr. Xiao Liu. The objective of the research is to propose a stochastic data-driven model that captures the complex interactions between the Interstate Highway System performance (e.g., crack propagation) and various critical environmental factors (such as temperature, traffic volume, humidity, precipitation, etc.). The research will enable a deeper understanding of how highway performance deteriorates over time under certain environmental/operating conditions, when the future highway performance will deteriorate to a critical threshold, and where/when preventive maintenance should be performed at highway sections.
During the Fall semester of funding, I experienced delays due to a high workload and COVID-19 limitations. Although these delays were unexpected, not all were bad. I was able to present a successful Fall 2020 Capstone Report. As capstone manager, I had to take extra time during the week to manage communication between stakeholders and set deadlines. Also due to COVID-19, my communication between me and my advisor was difficult, and I was not able to meet with him as I would like to. However, amongst these limitations I was able to become proficient with the programming language, R. As I continue my research, I am excited to learn more about data analytics and the processes that contribute to machine learning.
As I have mentioned before, I previously compiled potentially useful tables I found within the open-source dataset (LTPP) to identify each highway section with its respective data. Then I used parameters which were the survey date, transverse cracking length, latitude, longitude, precipitation, temperature, average annual traffic, and sections that have had maintenance or rehabilitation. My research will be determining which of these parameters affect highway degradation in a statistically significant manner. During this semester, I decided that using the programming language, R, was a better option than python. MissForest will then be used to impute missing values. In addition to learning basic commands, I also researched different machine learning models and their efficacy for this project. Different models I explored include SVM, K-Means Clustering, Hierarchical Association, ANN, Gradient Boosting, and Random Forest. I plan to test a few of my researched models to find the overall best model.
As a part of testing the models, I will also be looking at different types of highway response information. I will be extracting data to reflect alligator cracking in the highways, which are longitudinal cracks connected by transverse cracking. I will see if different response variables such as these are viable in a stochastic model for this application.
For the remainder of the semester, I plan to finish my research by creating the final model, establishing performance measures, and disseminating my report. Then, I will publish my results on Github so that future students or researchers can build upon my model and results. Overall, the predictive capabilities generated by the model will serve as the building block for more cost-effective preventive maintenance strategies. I plan to present my research at the Spring 2021 INFORMS Conference.