Measuring Forecastability of Time Series Data

[ggjx22]

Hello all, I'm in the midst of searching for an approach that is suitable to provide an indication that if an univariate time series data is forecastable or not. An indication can be a score or a measure, that I can infer from, that says that we can expect, when a model is trained on the data, the model is able to generalise well enough to produce accurate forecast. That is how I ended up in this package.

My searches have led me into several topics such as calculation of ADI and CV2, Approximate Entropy, MASE but none of them provide a clear guide on how to evaluate the result suitable for my use case.

My use case:

• Mix of regular and irregular time series in different scales and time periods
• Both regular and irregular time series are heterogeneous

Can anybody share their experience if this package has the features that is appropriate for my use case. Are there other better approaches?

I have tried time series clustering however I am unable to arrive at a satisfied conclusion because majority of my time series do not share common patterns.