Modeling Freight Vehicle Type Choice using Machine Learning and Discrete Choice Methods

Area of Science:

• Logistics and Supply Chain Management
• Transportation Science
• Machine Learning Applications

Background:

• Vehicle type selection is a critical logistics decision for firms.
• The process is complex due to multiple interacting agents and limited transportation data.
• Understanding these choices is vital for efficient freight operations.

Purpose of the Study:

• To model and predict road transport vehicle type choices for outbound shipments.
• To apply a random forest machine learning algorithm to capture complex interactions.
• To compare the performance of the random forest model against traditional logit models.

Main Methods:

• Utilized data from Commercial Travel Surveys on outbound shipment transportation.
• Employed a random forest machine learning algorithm for choice modeling.
• Calculated SHAP-based variable importance to identify key decision factors.

Main Results:

• Firm employment and shipment weight were identified as the most significant variables influencing vehicle type choice.
• The random forest model demonstrated superior predictive accuracy compared to multinomial and mixed logit models.
• Accuracy improvements were 7.8% over multinomial logit and 9.6% over mixed logit models.

Conclusions:

• Random forest models offer a more accurate approach to predicting vehicle type choice in logistics.
• Machine learning provides valuable insights into complex transportation decision-making processes.
• The findings can aid firms in optimizing vehicle selection for outbound shipments.