APBRmetrics

[mdelp]

Hello everyone. This is a great forum. I'm a Master's student in computer science and life long basketball fan. I managed to get a professor to agree to let me do a Data Mining project in the basketball realm.

I'm wondering if anybody has any good ideas for projects ideas. Some ideas I was thinking of follow:

I've read "A Starting point for Analyzing Basketball statistics" and the authors mention that one "unsolved" item in basketball is figuring out what it is about star players that makes them so effective and valuable despite the fact that their per minute numbers and shooting efficiency may not be as good as some of the top shooters. Intuitively, this has to do with creating their own shots, and shooting before the shot clock is up. A star player like Allen Iverson or Kobe Bryant will be able to create their own shots unlike a pinpoint shooter like Sasha Vujajic or Steve Kerr who needs to take an open assisted shot but can hit that with higher efficiency. Also, someone has to take a shot within 24 seconds. The ball is often put in the hands of the star player with a few seconds left on the clock so since everything else about the play has failed he will give them the best chance to score in the end. The idea of the project would be to formulate a hypothesis and prove it (i.e. that star players take more unassisted shots % wise, and make more % of shots then others with only a few seconds left in the shot clock), although this could probably easily be verified or falsified by looking at some the 82Games analysis.

So a better project would be to find any other features that separate out the star players, and/or correlate star players together with up and coming players, to help predict which players will be stars. For example do Rodney Stuckey's per-minute-stats indicate he will be a star one day because there are correlations with other stars?

Another problem that seems big and still "open" is that of determining the best defensive players from statistics. I know there are defensive efficiency measures out there, but I also know that it is a difficult problem since who one plays with can really influence their defensive plus - numbers. I'd like to look at this problem, but I'm not sure how I can add to what is all ready out there.

If you like my ideas, can make any recommendations on where to get statistics or what data mining techniques to use, know of anybody who has built a freely available play-by-play parser, or have some other good ideas to work on, please reply!

[MattB]

Something I have thought about from time to time is creating a central database and stat gathering app that would aggregate statistics from multiple sources...

Basically it would consist of a flexible interface that would allow for tracking very specific customizable events. If I wanted to track shots contested or potential assists (basically something more subjective or required human calculation) then I could use this interface to keep track of the player/time/position and so on.

This data would then get uploaded to a db and be a new data point that gets correlated to the game/team/player it's relevant to.

To make this a bit more feasible it might be worth tracking a few statistics (rather than opening it up to being user defined).

I think one of the most frustrating things for someone who does this as a hobby is the lack of availability to some of the more subjective stats. The benefit of this system would be that the few of us who are willing to take the time to keep track of these things would also benefit from others doing the same thing. Basically building a community of trackers.

I always thought it would be interesting anyway.

[Mountain]

If you wanted to look at the shot creation question it might be interesting to try to sort or even identify normal shots from so-called "created" shots. Put shots into a table by time on shot clock, floor location, game clock & situation and look at the FG%s.

If Kobe jacks up a 21 footer moving sideways is it a "created" shot? Created yes literally but is it created-good / neccessary? Does it matter if it with 17 seconds on the shot clock or 3? I'd say yes. Does it matter if it is down by 6 or up by 6? Yes. Hand in face or clean look? Yes. Teammates in better spots with less pressure? Yes.Break Kobe's shots into 20, 50, even 100 segments (for a season or as many as you can get data for) and look at the patterns. If he is creating good created shots exactly where, when and why is he doing so. Were his teammates cold or hot? Does he notice or care in early youth, championship years, after Shaq, now?

Check him and some others.


If you wanted to work on defense maybe correlate sum of overall counterpart defensive stats available at 82 games with 5 man lineup performance. How good are sum of average counterpart defensive stats at matching specific lineup defensive performance? Does a defender above or below average by x deviations cause an accelerated non-linear impact?

Or maybe assemble player level allowed shot charts of various kinds and then analyze them. Based on counterpart defense assumption or for the entire opponent with player on compared to off or the league average chart or player above average, average or below on counterpart data, adjusted +/- or team defensive rating. If a player looks above or below average by one or more of these where more specifically is this lean coming from?

[Harold Almonte]

Creating shots after pass is mainly a matter of an extra skill: ballhandling skill, and then physical advantage. Creators judge this advantage over their matchups and take decissions. I think an individual attempt with high success% is always better than lengthening the team off. poss. no matter the clock time, but studies tell us that extra passes increase FG%.

It would be interesting to look at the Mountain's idea of scorers's FG% or (points per poss.) on "not assisted-driving FGAs only" vs. rest of team's overall ppp., at different clock times, and see when extra passes are not a good option.

[thref23]