SSA - Singular Spectrum Analisys
Updated: Nov 18, 2019
We have some data. Our goal for today would be to find the cyclical pattern that the data presents.
Unfortunately, the number of `conversions` happen irregularly (we're not seeing each hour roughly the same amount of conversions). At the same time though, the `conversions` we're seeing have a clear pattern, one of which we've identified by looking at the charts is the day/night pattern. This makes sense: people usually buy during the day and not that often at night.
Obviously day / night is not the only pattern, there are potentially more, which look when summed up like complex looking shapes.
Our problem is identifying all these patterns, for:
tailoring our marketing campaing by spending more on the intervals with high conversion rate, and stop the campaing on the slughish intervals.
understand the main `conversion` driving factors for our clients (holidays, vacation periods, etc..?)
understand our clients. We've assumed day / night is a pattern but the cycles for days don't line up with our local time. This may be because:
our customers come from a different part of the world
it's not a day / night but a after-hours / morning-hours one
I hope I've convinced you that exactly knowing the characteristics of the patterns in our data is of great help.