Predicting customer sales

I would like to have your help on a topic of forecasting sales data of our customers.
I have a database where there are our customers, their turnover in each marketplace and the date of sale.
From this data I want to forecast the turnover of each customer in each marketplace for the next 5 months.
The workflow I have built does not allow me to make the forecast for the next 5 months and I have looked at knime Hub but none of the .

The history of my data is from January 1, 2020 to February 28, 2021.

I would like to have your help on a modeling method to do this.

Thank you for your help!

You could check out the collection of articles about regression models (not all of them are about time series):

The collection about time series might be especially interesting:

For future predictions you would typically have to provide a time series model that would accept future month and day dates (derived from historical data) or you might think about giving you historical data something like index numbers running from 1 … 1000 (or whatever your time period is) and then assign that to your prediction.

The question is always: would your model have some sort of seasonality (easter sales are different than christmas sales), sales in June are different from december (typically they are) and/or are they heavily influenced by outside events like your marketing campaign or advertisement spendings by your competitors. Let’s assume you are predicting mobile phone sales; they might look very different in a year (moth) when there is a new iPhone launch then at other times. You might have to insert a variable correcting that (attribution).

Or is there a monthly or weekly pattern (people buy right when they received their paychecks, or maybe at the end of the month when they still have money left).

Then from your data: do you want to predict the individual sales of single customers (you might then need a model for each of them, and you might not have that much data) or overall sales. It seems you want individual predictions; which is a further level of complexity.

What you always will have to ask yourself: is the reason driving the sales present in my data or not.

One example of such a sales prediction is the Kaggle “Rossmann sales” competition. The winner discusses some of what he did here:

One tool that promises to predict future time series is Facebook’s Prophet (I plan on writing a wrapper for KNIME for that but have not yet done so).

One example of such a prediction can be found here:

What you would be looking for is an “Out-of-Sample Forecast”. Of you want to learn more this book might be for you (I have not yet worked thru it):

For convenience one could try and bring some Python code into a KNIME node - which then would bring that back to a KNIME question.


Ah and then there is this blog article

And some time ago there was this example of the prediction of energy usage. You might want to explore the accompanying white paper and workflow

1 Like

Thank you very much for your answer. I think that with these methods I will be able to solve my problem


This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.