How to forecast a special day of the month


I’m struggling with forecasting the demand of the customers in a special day of the month. I have the daily data on the customers’ demand, weekend or not (0 and 1), first day of month or not (0 and 1), holiday or not (0 and 1). I used Arima models for forecasting and the result was ok with normal days of the month with very small diffirence between the real demand and the forecasted demand. However, on the first day of every month, due to the monthly promotion, the demand is significantly higher than that on other days. I tried to build a regression model as well. The prediction on the first day of month using regression was better than that using arima model. However, in general, R-squared was just 0.5 which is not high enough. Any one has any ideas which method I can use here to forecast the demand on the first day of the month most accurately? Many thanks!

Hi Jlmnguyen94,

You can correct the periodic fluctuations (seasonality) in your data using differencing. Particularly, from the current observation you need to subtract the value from the previous period (i.e. value of the observation a month ago). You can use the Lag Column node with the Lag value set to 30 to create lagged values and use the Math Formula node to subtract the the values. For more information please have a look at the example on the EXAMPLES server: knime://EXAMPLES/50_Applications/10_Energy_Usage/01_Energy_Usage_Time_Series_Prediction.


1 Like

Thank you very much Anna for your help. I will give it a try and hopefully it will work well.