what actually *is* SOTA for data modelling?

I have a pretty noisy tabular data set with 25 features. What method actually works the best for regressing on it? xgboost, neural nets, genetic algorithms or symbolic regression?

There is so much marketing it's hard to tell what actually works best.