Hello the problem is almost solved,
I realized there are missing points on my initial question:
First we wanted to leverage the power of ML Random Forrest algorithm in our Adobe Target
experience.
Therefore we wanted to use some insights we learned from our clients throught DMPs and find out
the discriminant values in thoses variables. But - using Global Mbox to fire up the experience we faced of
a problem :
Most discriminant variables were screen size, browser , or mobile / desktop information .
This was due to the fact that our current experience was far more efficient on mobile than on desktop,
This lead us to 3 conclusions :
- Machine learning was learning on wrong value
- Machine learning was eaysilly prone to platform impact
- Never setup a ML Experience without having the first tried an A/B Test
and went through analysis
So how did we solve the problem ? Easier than we thought but a little bit tricky also :
- Deploy a custom mbox with DTM (or any TMS)
- Send "choosen" parameters to feed the ML algorithm
- Refine the goal on which ML is learning from
More precise the goal is, more precise the result is !
In many way thinking that visiting page X should be the result
is not precise enougth !
By the way the part on Deploying custom mbox through DTM is de facto not
the easiest path to setup an industrialized means of production,
