Comparative analysis of Spatial-Hybrid Machine Learning Models for Vehicle Availability Prediction.

Tuesday, June 10, 2025

4:45 PM – 6:00 PM MT

Primary Author(s)

KT

Kelvin A. Terkper, GRA

Graduate Research Assistant
University of Louisiana at Lafayette
LAFAYETTE, Louisiana, United States

Presenter(s)

KT

Kelvin A. Terkper, GRA

Graduate Research Assistant
University of Louisiana at Lafayette
LAFAYETTE, Louisiana, United States

Co-Author(s)

MM

Milhan Moomen, PhD

Assistant Professor, Research, ITS/Traffic Research Manager
Louisiana Transportation Research Center
Lafayette, Louisiana, United States

Presenter(s)

MM

Milhan Moomen, PhD

Assistant Professor, Research, ITS/Traffic Research Manager
Louisiana Transportation Research Center
Lafayette, Louisiana, United States

Co-Author(s)

MR

M. Ashifur Rahman, PhD (he/him/his)

Research Associate V
Louisiana Transportation Research Center
Baton Rouge, Louisiana, United States
NM

Nurul-Haq Mohammed, GRA

Graduate Research Assistant
University of Louisiana at Lafayette
lafayette, Louisiana, United States
Wk

Waseem Akhtar khan, Masters

Graduate Research Assistant
University of Louisiana at Lafayette
lafayette, Louisiana, United States
JC

Julius Codjoe, PhD. & P.E.

Special Studies Research Administrator
Louisiana Transportation Research Center
Baton Rouge, Louisiana, United States

Submitter(s)

KT

Kelvin A. Terkper, GRA

Graduate Research Assistant
University of Louisiana at Lafayette
LAFAYETTE, Louisiana, United States

Abstract: Traditional techniques in vehicle availability forecasting, such as statistical and choice modeling (e.g., probit and logit models), have been widely used but often fall short in capturing complex, non-linear relationships within large datasets. This study investigates the integration of spatially-hybrid machine learning (ML) models to predict vehicle availability in Louisiana. The primary objective is to enhance the accuracy of vehicle availability forecasts by incorporating spatial variables into classic ML models. Techniques employed include a custom spatial random forest (SpatialRF) model for selecting spatial predictors, which were then used to train five ML models: Random Forest (RF), Artificial Neural Network (ANN), Support Vector Regressor (SVR), K-Nearest Neighbor (KNN), and Extreme Gradient Boosting (XGBoost). The models were evaluated based on their predictive performance, with the ensemble methods (XGBoost and RF) models demonstrating the highest accuracy. Key findings indicate that employment levels, household size, and income above the poverty line are significant predictors of vehicle availability. Shapley Additive exPlanations (SHAP) were used to interpret the importance of these variables, highlighting the benefits of integrating spatial analysis with ML. This research underscores the limitations of traditional statistical models in handling complex, non-linear relationships and emphasizes the potential of advanced spatially-aware ML techniques in transportation planning.

Learning Objectives:

Attendees can expect to learn the following from this session:

Upon completion, participant will be able to identify the most significant socioeconomic and spatial predictors influencing vehicle availability in Louisiana, such as employment levels and household size.
Upon completion, participant will be able to demonstrate how to apply ML techniques, including Random Forest and XGBoost, to improve the accuracy of vehicle availability forecasts by incorporating spatial variables
Upon completion, participants will be able to compare the predictive performance of different ML models used in the study based on metrics such as R², MAE, and RMSE.
Upon completion, participant will be able to explain the role of Shapley Additive exPlanations (SHAP) in interpreting the importance of variables within machine learning models used for vehicle availability prediction.