/sci/ - Science & Math

Retard here, or at least guy stuck with a problem that makes me feel like one.

Doing ML Regression; have a huge vector of like 3500 variables, which need to predict a single output value.

However, the error I'm seeing is absolutely terrible. Zero predictive ability from any of the input values, despite trying tons of different network topologies and setups.

Dataset is enormous.

Do I keep finding/trying to add more variables? What else can I try?

Is there some kind of meta-guide I can follow on how to troubleshoot this scenario?

Client is counting on me solving this and I would love to keep my job.