A Predictive LSTM Framework for Proactive Adaptive Traffic Signal Control

Urban traffic congestion necessitates a transition from reactive signal control toward proactive, prediction-driven traffic management strategies.This study proposes a multivariate forecasting framework based on Long Short-Term Memory (LSTM) neural networks to support short-term adaptive signal control at urban intersections.Using high-resolution traffic data collected at 20-second intervals, three independent yet structurally consistent LSTM models were developed to predict vehicle count, traffic density, and adaptive green time.The models exploit temporal dependencies through sequence-based learning and are trained using a supervised multivariate formulation.Experimental results demonstrate stable convergence and strong generalization, with validation loss values below 0.093 across all targets.Additional evaluation using RMSE, MAE, and MAPE confirms robust predictive accuracy under heterogeneous traffic conditions.Twenty-stepahead forecasts (approximately 6-7 minutes) reveal coherent temporal behavior, characterized by increasing and stabilizing traffic demand alongside converging green time allocations, indicating that the models capture key nonlinear interactions between congestion and control logic.Compared with conventional statistical and machine learning approaches, the proposed framework achieves competitive accuracy with lower computational complexity.The findings highlight the potential of LSTM-based forecasting to enable anticipatory traffic signal control, improve intersection performance, and reduce congestion-related environmental impacts.