Architectural Engineering PhD Defense
“A Computing Framework for Modeling the Functional Relationship between Electric Vehicle Uptake and Demographic Characteristics at a Granular Level throughout the U.S.” by Subhaditya Shom
9:30 am
Peter Kiewit Institute Room: 250
Target Audiences:
Contact:
Durham School, (402) 554-5935, durhamschool@unl.edu
Under the supervision of Dr. Moe Alahmad
Though electric vehicles have seen substantial growth in the last decade, this growth has been far from uniform geographically. In this dissertation a computing framework model is developed to analyze the functional relationship between the many quantifiable demographic factors which characterize an area at the ZIP code level and the EV uptake in those regions. This research fills an important knowledge gap of EV uptake, both by analyzing it at a finer geographic granularity, and by discovering the demographic factors which correlate with this growth. Heteroskedasticity and multi-collinearity exist for the independent and dependent variables, because of which new composite variables are generated using different constraints, such that the optimal level of correlation is achieved, using hypothesis testing and iterative feedback. Several established machine learning techniques are studied to develop a novel regression model for feature selection and prediction of EV uptake at the ZIP code level. The effects of these demographic features are quantified across 11 states in the U.S. The prediction model is then applied in the state of Nebraska as a test-case study, where the EV uptake is unknown at the ZIP code level. The XGBoost regression model is determined to be the best performing model with a 20-feature subset having an adjusted R2 of 0.88. To quantify the sensitivity of the regression model and learn about the output prediction variability, Shapley values are used.
Though electric vehicles have seen substantial growth in the last decade, this growth has been far from uniform geographically. In this dissertation a computing framework model is developed to analyze the functional relationship between the many quantifiable demographic factors which characterize an area at the ZIP code level and the EV uptake in those regions. This research fills an important knowledge gap of EV uptake, both by analyzing it at a finer geographic granularity, and by discovering the demographic factors which correlate with this growth. Heteroskedasticity and multi-collinearity exist for the independent and dependent variables, because of which new composite variables are generated using different constraints, such that the optimal level of correlation is achieved, using hypothesis testing and iterative feedback. Several established machine learning techniques are studied to develop a novel regression model for feature selection and prediction of EV uptake at the ZIP code level. The effects of these demographic features are quantified across 11 states in the U.S. The prediction model is then applied in the state of Nebraska as a test-case study, where the EV uptake is unknown at the ZIP code level. The XGBoost regression model is determined to be the best performing model with a 20-feature subset having an adjusted R2 of 0.88. To quantify the sensitivity of the regression model and learn about the output prediction variability, Shapley values are used.