Development of a traffic flow prediction and analysis model based on the Kolmogorov-Arnold Network (KAN) architecture

Abstract

This research study examines architectures and models for predicting and analyzing road traffic flows, with a focus on the Kolmogorov-Arnold KAN architecture. The study highlights the differences between this architecture and traditional machine learning architectures and compares their effectiveness. The main theorems and formulas of the KAN architecture are presented, and their theoretical foundations are explained. The results obtained based on the KAN model are reviewed during the study, and conclusions are drawn from these results. During the training process, the model showed 87% accuracy, while its accuracy in real-world predictions was found to be in the range of 87–90%. As a result, the researchers put forward the issue of further improving the KAN model and identified future development directions.