Integrating ADAM-based Neural Networks into Linear Optimization Models for Traffic Flow Prediction

Abstract

Urban traffic congestion particularly in rapidly developing cities poses further challenges on urban mobility management to keep pace with increasing demands on their existing infrastructure. A common approach to traffic flow analysis is to predict vehicle volume using different optimization techniques. This study proposes a hybrid traffic flow prediction framework that integrates neural network estimates based on Adaptive Moment Estimation (ADAM) method to linear optimization models. Historical vehicle counts collected during pre-and post-holiday periods where traffic congestion likely occurs were used in the method. By modeling traffic flow along critical intersections in road networks, the study also aims to assess the applicability of linear and quadratic programming–based optimization frameworks. Comparative analysis shows that our proposed optimization approach demonstrates better predicted and its results indicate the potential of the integrated ADAM-based predictions into the LP framework to improve traffic management and forecasting in urban areas.