Here's a better idea, don't use generalized statements but rather test your data first. If your models fits well (and doesn't overfit) after PCA, then go for i

Here's a better idea, don't use generalized statements but rather test your data first. If your models fits well (and doesn't overfit) after PCA, then go for it. If not, revisit. PCA has its place, and as the other commenter said, sure, it's not a feature selection algorithm. Or you can just feature select manually.