Evaluating the pedestrian level of service for varying trip purposes using machine learning algorithms

Area of Science:

• Urban Planning
• Transportation Engineering
• Human Factors

Background:

• Pedestrian comfort significantly influences the choice of sustainable travel modes.
• The Pedestrian Level of Service (PLOS) is a key metric for evaluating pedestrian facility quality.
• Understanding factors affecting PLOS is crucial for maintaining and upgrading urban walkability.

Purpose of the Study:

• To analyze the Pedestrian Level of Service (PLOS) based on work, education, and recreation trip purposes.
• To identify significant path and pedestrian flow characteristics impacting comfort.
• To develop predictive machine learning models for PLOS tailored to different trip purposes.

Main Methods:

• Data collection via pedestrian questionnaire surveys and sensors in Melbourne's CBD.
• Utilizing Mutual Information gain to select key influencing factors for each trip purpose.
• Developing and comparing Random Forest and Light-GBM machine learning models in Python.
• Employing SHAP (Shapely Additive explanations) for factor interpretability.

Main Results:

• Machine learning models achieved prediction accuracies of 0.74 (education), 0.80 (recreation), and 0.70 (work) using LightGBM.
• Key factors influencing PLOS varied across trip purposes.
• Interpersonal distance, proximity to vehicles, construction sites, vehicle volume, traffic noise, and footpath surface were identified as major influencers.

Conclusions:

• Trip purpose is a critical determinant of pedestrian comfort and influencing factors.
• Machine learning models provide accurate predictions of PLOS, aiding in targeted urban design.
• Findings support the development of more comfortable and sustainable pedestrian environments by addressing specific influential variables.