SCALABLE TRAFFIC FLOW ESTIMATION ON SENSORLESS ROADS USING LSTM AND FLOATING CAR DATA

Urban traffic monitoring is crucial for mobility, but the implementation of fixed sensors is costly and leads to restricted coverage. Floating Car Data (FCD) is emerging as an option, but its low penetration makes accurate traffic flow estimation difficult. This research proposes a Long Short-Term Memory (LSTM) model to scale FCD-based traffic estimates by learning flow patterns from routes with existing sensors. The model is trained with data from the most correlated sensors, but never the same one used for testing. The model identifies flow patterns from the available sensors and applies them to related paths. The findings indicate that the strategy is efficient on routes with consistent flow but has limitations in regions with high traffic variability. This work contributes to the advancement of FCD scalability methods, expanding the coverage of urban traffic estimation without the need for new infrastructure