I have a use case for traffic forecasting of various entities. Some of them have sinusoidal pattern and some others are frequent spiky events. I am looking for a technique that works for various types of traffic.
I have tried Informer transformer model (InformerForPredication, trained on my data). It works well with the sinusoidal patterns but does not perform well for the spiky traffics. The magnitudes of the forecasted spikes are much smaller (although they have regular cadence and present in the context).
Is there any other model that would work better for my use case? I’ll appreciate any recommendations.