Prediction Framework for Searching for Parking Average Vehicle Cruising Time and Emissions in Urban Areas

This thesis proposes and implements a novel framework to identify the factors influencing average cruising time (ACT) related to searching for parking and uses multiple machine learning (ML) models to predict ACT and average vehicle emissions (AEM) in New York City (NYC) and Los Angeles (LA). Specifically, the author 1) analyzes NYC datasets by spatial lag models to explore the factors affecting ACT; 2) uses the K-Means method to cluster land use type and vehicle composition, and perform a comparative analysis for different groups; 3) utilizes ML models to predict the land use and vehicle composition in NYC and LA, and use them into a Bayesian Ridge regression model to predict the ACT; and 4) uses a Gaussian-Link generalized linear model to calculate average cruising distance (ACD) based on ACT and other variables to estimate AEM generated by parking events. The major findings include 1) Residential and commercial areas have a significant positive correlation with ACT. The parking hotspots roughly coincide with the distribution of density, land use, and job opportunities, which may be associated to growing residential freight demand and sluggish parking supply. 2) A parking hotspot with high ACT is not necessarily a location with high AEM. Compared to locations with high ACT in the central areas, the heavy-duty-truck- (HDT)-dominated areas at the periphery of the city can generate higher vehicle emissions. 3) AEMs are more affected by the type of dominating vehicles in the grid, and HDT-dominated grids generate four times of pollutants than light-duty-truck (LDT)-dominated ones