Predicting High Temporal Deviation in Traffic Flows Based on SUMO Simulation

Bakbergen Mendaliyev, Yedilkhan Didar

2025

Closed access Metadata (CSL-JSON)
DOI 10.1109/sist61657.2025.11139163
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

The study aims to predict high time deviations (HighLoss) in traffic flows based on SUMO simulation data. Machine learning methods are used to solve this problem, including logistic regression, decision trees, SVM, random forests and gradient boosting. Model optimization is performed using GridSearchCV, which increases classification accuracy and reduces the risk of overfitting. The main evaluation metrics are Accuracy, Precision, Recall and F1-score, which allow for a comprehensive analysis of the performance of each algorithm. Ensemble methods (Random Forest and Gradient Boosting) achieve Accuracy of 98 percent and demonstrate more balanced Precision and Recall indicators compared to SVM. At the same time, Gradient Boosting has Recall of 96.4 percent, which is slightly inferior to Random Forest with 96.8 percent. At the same time, SVM shows Accuracy of 99.5 percent, but is inferior to other models in Recall. The most important characteristics affecting the prediction of time delays are trip duration, waiting time and route length. The experimental setup, based on Python and SUMO, provides controlled conditions for testing the models. The main limitation of the work remains the use of modeling data, which requires additional testing of the algorithms on real transport systems. Further work will focus on expanding the set of features taking into account weather conditions and congestion, as well as integrating the proposed solutions into real road networks.