The main thing in ML is not a algorithm of ML but finding the correct features with correct normalization, and usually this takes most of the time.
On Kaggle there are many topics that shows how fixing normalisation of one feature can lead to lowering error rate from 40% to 5–10%.