Urban traffic congestion remains one of the most pressing challenges for modern cities, affecting the overall quality of life. This study aims to identify and classify traffic states on two central roads in Athens, Greece, using a multi-stage approach that combines clustering, classification, and explainable AI techniques. Traffic speed data were collected from the Google Directions API, while hourly load data were obtained from detectors operated by the Traffic Management Centre of Attica. Covering the period from January to July 2022, the datasets were combined spatiotemporally, with a focus on peak periods during both weekdays and weekends to capture high-demand traffic conditions. K-means clustering was applied to identify two primary traffic states: (i) congested, with higher load and lower speeds, and (ii) less-congested, with lower load and higher speeds. To mitigate class imbalance, the SMOTE oversampling technique was employed. Four XGBoost classification models were trained separately for each road with two directions and evaluated using confusion matrices and standard performance metrics. SHAP values were then used to interpret model predictions and assess feature importance. Weekday, hour during the day, and temperature emerged as the most influential variables. These findings offer valuable insights to support urban traffic management and planning strategies.
| ID | pc671 |
| Presentation | |
| Full Text | |
| Tags |
