MAPE

[lilailai688]

The MAPE on the PhysioNet dataset is high, and generally, a MAPE above 50% indicates poor predictive performance of the model. Is it reasonable to set up this irregular time series forecasting task?

[willzhang3]

We did not use MAPE in this paper, as we find MAPE is not well-suited for evaluating this task. This is due to the normalization of time series values to the range [0, 1], resulting in some denominators in MAPE that are very small or nearly zero. The prediction error in terms of these small values will dominate the MAPE, making this metric overly sensitive and unstable.