Author ORCID Identifier



Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Lane Department of Computer Science and Electrical Engineering

Committee Chair

Natalia Schmid

Committee Co-Chair

Arvind Thiruvengadam

Committee Member

Gianfranco Doretto

Committee Member

Xin Li

Committee Member

Vishnu Padmanaban


One of the major contributors of human-made greenhouse gases (GHG) namely carbon dioxide (CO2), methane (CH4), and nitrous oxide (NOX) in the transportation sector and heavy-duty vehicles (HDV) contributing to about 27% of the overall fraction. In addition to the rapid increase in global temperature, airborne pollutants from diesel vehicles also present a risk to human health. Even a small improvement that could potentially drive energy savings to the century-old mature diesel technology could yield a significant impact on minimizing greenhouse gas emissions. With the increasing focus on reducing emissions and operating costs, there is a need for efficient and effective methods to predict fuel consumption, maintenance costs, and total cost of ownership for heavy-duty vehicles. Every improvement so achieved in this direction is a direct contributor to driving the reduction in the total cost of ownership for a fleet owner, thereby bringing economic prosperity and reducing oil imports for the economy. Motivated by these crucial goals, the present research considers integrating data-driven techniques using machine learning algorithms on the historical data collected from medium- and heavy-duty vehicles. The primary motivation for this research is to address the challenges faced by the medium- and heavy-duty transportation industry in reducing emissions and operating costs. The development of a machine learning-based approach can provide a more accurate and reliable prediction of fuel consumption and maintenance costs for medium- and heavy-duty vehicles. This, in turn, can help fleet owners and operators to make informed decisions related to fuel type, route planning, and vehicle maintenance, leading to reduced emissions and lower operating costs. Artificial Intelligence (AI) in the automotive industry has witnessed massive growth in the last few years. Heavy-duty transportation research and commercial fleets are adopting machine learning (ML) techniques for applications such as autonomous driving, fuel economy/emissions, predictive maintenance, etc. However, to perform well, modern AI methods require a large amount of high-quality, diverse, and well-balanced data, something which is still not widely available in the automotive industry, especially in the division of medium- and heavy-duty trucks. The research methodology involves the collection of data at the West Virginia University (WVU) Center for Alternative Fuels, Engines, and Emissions (CAFEE) lab in collaboration with fleet management companies operating medium- and heavy-duty vehicles on diesel and alternative fuels, including compressed natural gas, liquefied propane gas, hydrogen fuel cells, and electric vehicles. The data collected is used to develop machine learning models that can accurately predict fuel consumption and maintenance costs based on various parameters such as vehicle weight, speed, route, fuel type, and engine type. The expected outcomes of this research include 1) the development of a neural network model 3 that can accurately predict the fuel consumed by a vehicle per trip given the parameters such as vehicle speed, engine speed, and engine load, and 2) the development of machine learning models for estimating the average cost-per-mile based on the historical maintenance data of goods movement trucks, delivery trucks, school buses, transit buses, refuse trucks, and vocational trucks using fuels such as diesel, natural gas, and propane. Due to large variations in maintenance data for vehicles performing various activities and using different fuel types, the regular machine learning or ensemble models do not generalize well. Hence, a mixed-effect random forest (MERF) is developed to capture the fixed and random effects that occur due to varying duty-cycle of vocational heavy-duty trucks that perform different tasks. The developed model helps in predicting the average maintenance cost given the vocation, fuel type, and region of operation, making it easy for fleet companies to make procurement decisions based on their requirement and total cost of ownership. Both the models can provide insights into the impact of various parameters and route planning on the total cost of ownership affected by the fuel cost and the maintenance and repairs cost. In conclusion, the development of a machine learning-based approach can provide a reliable and efficient solution to predict fuel consumption and maintenance costs impacting the total cost of ownership for heavy-duty vehicles. This, in turn, can help the transportation industry reduce emissions and operating costs, contributing to a more sustainable and efficient transportation system. These models can be optimized with more training data and deployed in a real-time environment such as cloud service or an onboard vehicle system as per the requirement of companies.