A supervised machine learning model for imputing missing boarding stops in smart card data

Area of Science:

• Transportation Science
• Data Science
• Urban Planning

Background:

• Public transport is vital for urban mobility, generating vast smart card data for travel behavior analysis.
• Data integrity issues, such as missing boarding stop information, hinder accurate analysis.
• Existing methods struggle with incomplete public transport datasets.

Purpose of the Study:

• To develop a supervised machine learning method for imputing missing public transport boarding stops.
• To introduce a novel evaluation metric, Pareto Accuracy, for ordinal classification tasks.
• To assess the method's robustness, generalizability, and performance against existing imputation techniques.

Main Methods:

• Utilized a supervised machine learning approach based on ordinal classification.
• Integrated General Transit Feed Specification (GTFS) timetable, smart card, and geospatial data.
• Developed and applied a new metric, Pareto Accuracy, for evaluating ordinal imputation algorithms.

Main Results:

• The proposed method accurately imputes missing boarding stops, outperforming traditional imputation techniques.
• The approach demonstrates robustness to irregular travel patterns and does not require additional data mining.
• Transfer learning validation confirmed the model's generalizability across different urban contexts.

Conclusions:

• The developed machine learning model effectively addresses missing data in public transport systems.
• The Pareto Accuracy metric provides a reliable evaluation for ordinal classification problems.
• This research offers significant implications for enhancing transportation planning and travel behavior research through improved data quality.