Share result for Mini Data Hack | Forecasting



Please share your approach for the Mini Data Hack , where the objective was to predict demand in next 1 year…



I’ll add my approach but with a very significant caveat that I only ranked around 50th percentile on the public board. I am relatively new to this so I am highly interested in hearing other, more successful approaches.

A quick plot of sales qty over time showed a couple patterns. There appeared to be certain days of the year with unusually high spike in sales volume. Also, there were certain dates each month which often had sales volume which appeared higher than normal. Perhaps these coincided with holidays or promotions. As everyone knows, seasonality plays a large part in forecasting retail sales.

Since there was only 1 independent variable to work with (date) I decided to split it into separate parts: year, month, date. I threw those 3 variables into a linear regression model to get my predicted output. I thought about adding some sort of multiplier to account for year over year growth but figured the year variable would already do that.

That’s really all I had time for. I’m using Python and still a beginner with the language (professionally use SAS, SQL). So everything takes longer…

Looking forward to hearing others…