I am working on a sales forecasting model. I have got the transactional data of 50 stores of an organization. Each transaction has been recorded for each store on daily basis. So there are multiple entries for one store and a given day. The data I have is from Jan 2016 to Feb 2018. Now I need to design a corporate sales forecasting model/dashboard for higher management. So that they can compare the sales on daily basis with the same day on last two year with the forecasted sales of 2018, to see the growth. (I mean if they select 2nd Wednesday of April, it should show the sales of 2nd Wednesday of April 2016, 2017, along with 2018(forecasted one).
Tools I am using here is R, Power BI.
So can you please tell me how should I approach this problem. I will be very thankful to you all.
I can try and help to get you started. Will you be predicting all stores together or separately? If together just sum up the transactions across all stores and have a count for each day. Otherwise, you will have to run each store through your model independently.
I would try a number of models first, optimize them, and see which one is giving you the lowest error. Some model ideas given the details you've provided include:
nnetar- feedforward neural network- H
auto.arima - H
prophet (facebook algorithm) - H
All the models above can be found using R- the forecast package and prophet is a separate package. If there is an H next to the model, that means the model can account for holidays- so you can add a covariate- which you will likely want/have to do. Holidays will affect your data and you want your model to account for them.
You will have to make sure you read the documentation for each model, so you perform your analysis correctly and can optimize them.
Make sure you have train and test date and calculate error to compare models and which is performing better. You may also want to consider an ensemble model. This is something you can google for more information.
That should get you started.
Thank you so much for the response. I really appreciate your help. I don't have a column yet for holidays but yes It will be good if I can have one. I will try to add one on the basis of a weekday. I have added the transactions in Pivot and got the daily sales for each store. I want to forecast sale for all the stores for next 6 months. I have attached the dataset for 2017, please have a look at it and can you please elaborate it little more.