Deep Learning for Road Traffic Forecasting

Unpredictability

Traffic forecasting faces significant challenges due to the inherent unpredictability of traffic behavior. Eric Manibardo highlights that while short-term predictions rely on recent traffic data, long-term forecasts must consider various factors like holidays or nearby events 1. He emphasizes that traffic behavior is complex and influenced by numerous unpredictable events such as accidents or sudden stops 2.

We are not going to hit never the 100%. It is impossible because the traffic behavior as a natural task, it is very, very complex.

--- Eric Manibardo

Understanding these factors is crucial for improving the accuracy of traffic models.

   

Data Quality

Data quality is a critical issue in traffic forecasting, with challenges arising from noise and incomplete data. Eric Manibardo notes that GPS data often has mediocre quality due to positional errors and limited vehicle type coverage 3. He suggests focusing on traffic profiling and graph neural networks to enhance data modeling and prediction accuracy 4.

Mainly data preprocessing, without a doubt. Because as you said, you need a vast amount of traffic data.

--- Eric Manibardo

Effective data preprocessing is essential to mitigate these issues and improve forecasting outcomes.