Application of Machine Learning to Child Mode Choice with a Novel Technique to Optimize Hyperparameters

Area of Science:

• Transportation Science
• Machine Learning Applications
• Child Mobility Studies

Background:

• Travel mode choice (TMC) prediction is critical for transportation planning, yet research on children's TMC is limited.
• Existing studies primarily focus on adult TMC and home-to-school trips for children.
• Predicting children's non-school related TMC requires advanced analytical methods.

Purpose of the Study:

• To predict children's travel mode choice (TMC) using machine learning.
• To identify key determinants influencing children's TMC.
• To introduce and evaluate a novel multi-objective hyperparameter tuning (MOHPT) technique for machine learning models.

Main Methods:

• Light Gradient Boosting Machine (LGBM) was employed for TMC prediction and determinant analysis.
• A novel Multi-Objective Hyperparameter Tuning (MOHPT) technique was developed and compared against random search, grid search, and Hyperopt.
• Logistic regression was used to interpret the influence of identified parameters on children's TMC.

Main Results:

• MOHPT demonstrated superior performance over conventional methods in hyperparameter tuning, balancing prediction accuracy and computational cost.
• Key determinants of children's TMC include trip distance, origin walkability and bikeability, child's age, and household income.
• LGBM successfully identified influential variables for children's travel behavior.

Conclusions:

• The proposed MOHPT technique enhances machine learning model performance for children's TMC prediction.
• Children's travel mode choices are significantly influenced by environmental factors (walkability, bikeability) and socioeconomic variables.
• Findings support promoting sustainable transportation options for children by considering these determinants.