also in many cases, ML and consorts are mostly used to compensate for shitty data. So if you are the one acquiring data, do it well in the first place. You don't (usually) need big data you need meaningful data.