Multilayer Perceptron-Based Traffic Signal Timing Optimization for Improved Urban Intersection Performance

Abstract

Urban traffic signal systems based on fixed-time plans may not adequately respond to variations in traffic demand among signalized intersections, limiting their operational efficiency. This study proposes a data-driven framework based on a Multilayer Perceptron (MLP) to support traffic signal timing optimization in an urban corridor. The framework uses historical traffic observations to estimate traffic volume, cycle length, and effective green time and subsequently transforms the predicted outputs into feasible signal-timing configurations while preserving the existing two-phase structure. The approach was evaluated using 5880 intersection–interval records collected from six signalized intersections and compared with the existing fixed-time configuration. The results showed consistent operational improvements across all six intersections. The Vehicle Congestion Index (VCI) decreased by 23.7–65.6%, while average control delay decreased by 9.5–26.2%. The largest improvement was observed at Intersection 2, where the VCI decreased by 65.6 and average control delay decreased by 26.2%. The MLP-based configurations also modified cycle length, green time, and red time according to the operating conditions of individual intersections rather than applying a uniform adjustment. Statistical analysis supported the observed reduction in control delay, with a paired-samples t-test yielding p = 0.000782 and a Wilcoxon signed-rank test yielding p = 0.015625. These findings indicate that integrating traffic prediction with signal-timing optimization can provide a practical data-driven alternative to conventional fixed-time control, particularly in environments where fully adaptive traffic-control infrastructure is limited. The framework was evaluated offline within a single six-intersection corridor; therefore, further validation using real-time data and broader networks is required.