Integrating Signal Performance Data for Turning Movement Counts Estimation at Intersections Using Machine Learning

This paper explores the use of signal performance measure (SPM) data to estimate turning movement counts at intersections. Accurate turning movement counts are essential for signal timing optimization and traffic management, but traditional methods can be costly or limited in coverage. A novel approach using SPM data is proposed to alleviate the challenges associated with obtaining accurate traffic counts on approaches with traditional detection not designed exclusively to gather such data. The research objectives include obtaining SPM and turning movement counts, comparing various machine learning models to identify the most effective approach, testing the model on a new data set, and analyzing potential limitations and improvement areas. The methodology involves data collection from an intersection, processing this data for model development. Machine learning models were explored, including decision trees, linear regression, random forest, K-nearest neighbors, and extreme gradient boosting (XGBoost). After evaluation, XGBoost was found to have the highest R2 value of 0.754 and a root mean squared error of 5.44, demonstrating reliable performance. Future work suggests enhancing the model with additional data types, exploring hybrid models, and adapting the model to different urban contexts. This research contributes a new methodology to the field, offering practical insights for creating a more efficient and effective model for estimating turning movement counts.