Advice on Deep Learning for store sales

I understand this question is beyond the scope of this class, but wonder if anyone could provide some insight…

Are there any papers or pointers to work done on how to effectively apply deep learning to problems like store sales forecast?

Store sales forecasting has at least two levels of seasonality (month and week). In addition, there may be yearly trends as well.

The practical example adopted in week 4 of this class (sunspots) has seasonality, but no yearly up or downward trends. Even so, by combining multiple tools (conv1d, lstms, dense, etc), it was still far from ideal.

Based on my naive beginner knowledge, it seems the best tool out there is Facebook Prophet, which as fas as I understand is not based on deep learning. It adequately fits predictions on human-scale series. Plus, it can be fine tuned with the notion of holidays, for instance.

So I’m not asking for a tutorial here, but rather simply if anyone can provide any pointers to any papers or any work out there that gets results with deep learning at least as close to what Facebook Prophet, for problems like store sales forecasting.

Thank you,

Julius Lerm

That’s quite a broad topic and agree not all need to be done using Deep Learning… :slight_smile:
As usual it depends on your data, what you are looking for- for instance deep learning models can be used to generate forecasts for new SKUs with little to no historical sales data, time and budget.

I’ve read this article that explained some other approaches, from traditional to Deep Learning:

Amazon’s white paper is also interesting - I’ve skipped how to implement and focus on rationals:

Maybe it would be a good exercise to compare what Facebook provides to what you can get using other techniques - see how can be done here:

Hope it helps,


Thanks a lot!