Evaluation of Traffic Flow Prediction Models in a Real-World Urban Intelligent Transportation System Pilot

Accurate traffic flow prediction is a critical component of modern Intelligent Transportation Systems (ITS), supporting real-time decision-making and congestion mitigation in smart cities. This paper presents a comparative analysis of two different approaches to traffic flow prediction problem: a temporal Transformer-based model (iReformer) and a spatio-temporal Graph Convolutional Network combined with Long Short-Term Memory (GCN-LSTM). The evaluation is conducted using a real-world urban traffic dataset from Istanbul, Turkey. The models are assessed under several different prediction settings, varying in historical input window length and forecasting horizon. Results show that GCN-LSTM performs best in short-term prediction scenarios, while iReformer achieves better performance with longer input sequences. According to the authors best knowledge, this is the first comparative analysis of novel temporal and spatio-temporal approaches on real-world urban data.