In my first post on the Big Data Cup, I talked about how I think I’m going to focus more on the Scouting dataset. This post will talk about how I will approach scouting a player.
When thinking about a GM’s role, I’d imagine that there are two competing desires when drafting a player: what’s the best player available at the moment, and what need do I need to fill at the moment? The first approach would be an NBA “Drafting the best talent” approach, while the second approach would be based more on need.
By perusing Cleaning the Glass, Ben Falk and other contributors point several times that players are useful in specific situations. Instead of drafting a center in the NBA, you could draft a stretch three-point shooter, a strong rebounder and defender, etc. All of those players could serve different needs, depending on what the rest of one’s team looks like at the moment.
My approach to the hockey scouting approach will be similar. Roughly breaking the players into “offensive” and “defensive” players, I’d imagine that individual players have strengths: long-distance scorer, high-efficiency shooter, etc. Similarly, the defensive players may be good at certain things - steals, blocked shots, etc. My play is to do some introductory analysis on different types of players to get an idea in my head of what results could look like, and then programmatically approach how to cluster players similarly.