How can we tell a machines model represents a good likeness of the "real world" ?
Something to consider when developing a strategy or machine learning model are.
What does the original data set look like / come from?
How were missing cases and values dealt with?
Finally in Experimental design "It is wise to take time and effort to organize the evaluation properly?
Ensuring that the right type of data, and enough of it, is available to answer the questions of interest." Ref More..
Using a balance of statistical and machine learning methods tends to give the best performance overall.