Improve holtWinter forecast model accuracy



Hello, Currently i’m using Holt winter model for daily sales forecasting. I have two different type of data. One is restaurant data and another one is lifestyle store data.

When I’ve forecasted restaurant data and match with actual data then prediction is approximately same but same model not work on lifestyle store data, the difference between actual sell and predicted sell is very large for lifestyle store data.


The best way to correctly use any time series algorithm is by plotting both data and it’s predictions. Try varying the coefficients and notice their impact on the predictions. It may be possible that the coefficients you use for one dataset may not work for other. Try varying them and when it’s seems that the predictions are close to the actual values, then use that model for forecasting.



Currently i’m running single forecasting model (Holt winter) on both data.
I want to create a forecasting model in which model automatically detect holiday, weekdays,weekends.
Is it possible ??
If yes then how ?? :smiley:



It seems more like a classification problem rather than a Time series. Do elaborate if I am wrong.