I do not know why you considered sales forecast are not belong to Machine learning.
Sales forecast is a prediction technique, and historical data is required to predict future sales.
It can be done with ML, but it has some peculiarities (like day of the week effect, day of the year effect) which are harder to capture by ML models.
There are ways to do that, but think about Chrismas sales, Black Friday sales. Periods like the FIFA World Cup, the Olympics. In muslim countries – Ramadan.
As of now, all these effects cannot be quantified accurately by ML algorithms. That’s why we were shy to put it there.
Of course, ML can greatly help predict who can buy and when.
For instance, I manually make the analysis/strategy of Christmas sales. When a person comes to my website, I can use an ML algo to predict if he is going to buy. That’s ML forecasting this sale, but how I got this person to the website is a different thing.
One factor to think about is advertising. You think you can spend X money to bring a client to your website around Christmas, but thousands of other companies are bidding for ad space, too.
Further, let us quote a student of ours:
‘Adding from my own experience, multi-department sales forecasting is used a lot in machine learning. In fact, Walmart and other stores have set up Kaggle competitions in an effort to see how people do, which also helps them find great data engineers and machine learning developers.
Basically sales forecasting are time series forecasting problems and machine learning can be quite effective at these. The techniques used are multi-step forecast or sequence forecast.
In that capacity, “day of the week effect, day of the year effect” can be incorporated into your models. The bigger problems are usually data set imbalances. In these contexts, time-based demand forecasting will usually be paired up with cross-sectional classification methods.
Regarding the multi-department sales example, and coupling that with temporal considerations, how this is generally done is you look to see if there are spikes in sales in similar times throughout the year. You can then check if they correspond to holidays or events. Then you can check what proportion of departments had spikes in sales at around those same time frames. You plot the sales of different departments and you can then learn a lot about forecasting sales. For example, some departments (clothing, sports, etc) are impacted by certain holidays, events, times of year while others are not. Seasonal decomposition is also applied to machine learning to basically understand the impact of seasonal events (recurring and otherwise).
One of the more popular examples of sales forecasting were the Rossmann store sales. Machine learning algorithms were able to forecast six weeks of daily sales for over one thousand stores that were regionally distributed.’
Hope this helps!
The 365 Team