completed
Traffic Flow Forecasting
Predictive models comparing LSTM, XGBoost, ARIMA, and Prophet for vehicular traffic-flow prediction.
Overview
Research work at IASBS comparing four forecasting approaches on vehicular traffic-flow data, to see which handled the sharp, recurring peaks of rush-hour traffic best.
Comparison
| Model | MAE | Handles seasonality |
|---|---|---|
| ARIMA | moderate | manually specified |
| Prophet | moderate | automatic |
| XGBoost | low | via engineered features |
| LSTM | lowest | learned implicitly |
Takeaway
LSTM captured the sharp rush-hour peaks most accurately, but at a real cost in training time and tuning effort — XGBoost with well-chosen lag and calendar features came close at a fraction of the compute, which matters when the model needs to be retrained regularly on fresh data.