In my experiment(train 4 years to predict 1 year, to predict 2015~2020), using the future data to normalize dataset(use the maximum of testset) will significantly improve the result. In another reply, you said you got the similar results by different normalization methods, this is very confusing to me, have you tried data from other years?
Thank you very much.
fulifeng
commented
2 years ago
In my experiment(train 4 years to predict 1 year, to predict 2015~2020), using the future data to normalize dataset(use the maximum of testset) will significantly improve the result. In another reply, you said you got the similar results by different normalization methods, this is very confusing to me, have you tried data from other years? Thank you very much.