Quantifying Correlated Variables and Determining Signal Phase Reservice Rate Models from Multiple Linear Regression Using Historical Signal Controller Data
Loading...
Authors
Reyes, Marcus Aldrin Acorda
Issue Date
2025
Type
Thesis
Language
en_US
Keywords
Controller , Reservice , Signal Timing , Traffic , Traffic Operations
Alternative Title
Abstract
Signal phase reservice is a technique that operates in coordination under advanced controller settings that help mitigate unreasonable delays at side streets when the green dwells on the main street. Despite its potential benefits, evaluating the effectiveness of phase reservice remains challenging due to its reliance on in-field observation, which is time-consuming and labor-intensive. Its current implementation also depends heavily on local expertise and engineering judgment. As a result, the practical impact of this technique often remains unclear. This study introduces a data-driven approach to evaluate phase reservice using historical controller data from eight intersections in the Reno-Sparks region. Data from Cubic Trafficware’s ATMS, including Split History, Timeline Split History, and Occupancy Reports for March 2025, were analyzed to identify instances of phase reservice and the corresponding traffic conditions. Through a regression analysis, the data can confirm engineers’ initial assumptions and create a guideline to implement phase reservice in coordinated intersections. A linear regression analysis was conducted to investigate the relationship between reservice rate and four variables: coordinated occupancy rate, non-coordinated occupancy rate, number of phases, and average cycle length. The analysis found significant correlations for all variables except coordinated occupancy rate. A multiple linear regression model incorporating all four variables yielded a residual standard error of 0.2236 and an adjusted R² of 0.4697. By plotting the actual vs estimated reservice rates, it was deemed that creating an accurate model was infeasible with current data. To address this, a regression decision tree was developed, achieving a cross-validation error of 0.397 with seven splits. This model is a more practical tool that can be used by engineers to help facilitate decision making in addition to engineering judgment. Future research should explore more diverse intersections and advanced modeling techniques to further enhance predictive performance.