I am giving a presentation on Data Science, and I want to talk about the idea that data that is not "big" enough is a big barrier for Machiene Learning. Looking online, there are concepts like overfitting and underfitting, but I am more looking to talk about data that, even if fitted optimally, would still not actually be a good model for the system.
Is there a good term to use for this?
Small sample size is probably the concept you're looking for. A common failure in statistics is trying to draw conclusions from data that isn't "big" enough to accurately represent the underlying distribution.
It's worth noting that "small data" increases your chances of overfitting - which is another way of saying that your model is weak to outliers and noise. It's also worth noting that bootstrapping, simulation, and duplication might help (but not always).
There may be some other niche term for this, but if you relate it to basic stats there's a high probability everyone in the room will understand what you're talking about.
