In terms of actively validating the model, we assessed the validity of our results, as well as fine tuned our computation and aggregation models. Based on the results of this validation, we decided to benchmark per position and use a two-year time frame since this provided the most preferable sample size in playing time. During this time frame, the player needs to have played at least 30 minutes in 10 matches.

By benchmarking players per position and league, the insights that can be derived are relative to others in the same category. In sum, making player comparison in these metrics more appropriate.

When it comes to accuracy, the model returns very impressive results for an event data-driven methodology.

You can read the full validation study here.