Fig. 8: Mean absolute error for various choices of (s, f).
From: Data-driven analysis and forecasting of highway traffic dynamics
a, b The top row was generated utilizing the MH-DMD algorithm. c, d The bottom row was generated utilizing the DMD algorithm. It is evident that the Hankel-DMD can provide greater accuracy than a standard DMD analysis. This is best seen by observing the increase of blue-colored squares in (a, b) versus (c, d). In both cases, the results suggest that the forecasting of traffic is a temporally local task and that current and future conditions depend very little on the historical past. Furthermore, increasing the amount of historical data utilized can hinder the accuracy of forecasts. This demonstrates the need for intelligent transportation systems to have sufficient analytic capabilities of detecting dynamic traffic patterns in real-time and that training over extensive amounts of historical data is not necessary.
