How to predict which is the game time for each player

Home for all your discussion of basketball statistical analysis.
Post Reply
josecarlos
Posts: 67
Joined: Sun Oct 08, 2017 10:11 pm
Location: Spain
Contact:

How to predict which is the game time for each player

Post by josecarlos » Mon Mar 19, 2018 9:19 pm

Hi!!!

I'm gonna try to explain an idea the best way that I can with my english :) I have been thinking how to detect which is the best five starter based on statistics respecting the structure of a team, guards, forwards and center.

How to measure each player we can do it on base a parameter which we can call it "benefit". This parameter can be calculated from the boxscore of each player as it is already done in some fantasy leagues, for example. So, for each player we've got their total benefit in a season, the benefit per game and minute. It is easy to select the best five players to start the game. The best five players with better benefit by minute respecting the structure of a basketball team. But ... the players has the bad habit of getting tired :) and we have to make changes.

Everything is clear, isn't it? So, what I want to do is to make an algorythm that select this five starter, this is easy, and the respective changes and when to do these changes, always getting the best benefit team on the floor. I have been thinking in doing a variation of knapsack algorythm, but right now I don't know how to do it. Maybe someone of you know a better algorythm or have any other idea to select the players or maybe you consider that this idea is useless.

All your comments are welcome!!!

Crow
Posts: 5557
Joined: Thu Apr 14, 2011 11:10 pm

Re: How to predict which is the game time for each player

Post by Crow » Wed Mar 21, 2018 4:53 pm

If you want best overall team results, you need to know the kinds of "benefits" and how players help each other with shooting, passing, rebounding, and drawing / giving fouls (the "factors") and how they compete to provide these benefits / duplicate each other. It is not a simple summation of benefits. Perhaps a knapsack algothrim will better capture the dynamics at work than a simple summation.

To the extent that coaches think it is a simple summation and they can just throw players with their known benefits on the court any which way and get that expected return, they reveal themselves as not adequately analytic or analytically well informed. Even beyond small sample randomness, some players "fit" better than others.

One source of evidence of overlap, competition, and bad fit or good fit with greater than average simple linear returns of benefits: viewtopic.php?f=2&t=9459

In NBA does RPM better predict lineup performance than box score metrics? I have assumed so, but we need a detailed analysis of each alone and each possible blend of the two (or more tha two). RPM tries to capture overall impact but some players are better in sone situations than others. (So we NEED imo RPM factors and splits. The most logical ones over NOT looking / considering them. And then we need GMs and Coaches committed to disciplined testing, " truthing" better than whatever is done now by eye or simple summation and then vigorous optimization of lineups.)

Fastbreaks benefit from defensive rebounds. Jumpshots benefits from good passes. Offensive rebounds benefit from rebound opportunities. Dunks and open 3s benefit from open opportunities coming from offensive rebounds, drive n dish, double teams, creative passes. Getting to line benefits from space. Fouling can stop a defensive breakdown or discourage a drive. Or cost points or interfere before help arrives. Etc. Could look at evry possible interactive impact or focus on the "factor" level impacts.


josecarlos
Posts: 67
Joined: Sun Oct 08, 2017 10:11 pm
Location: Spain
Contact:

Re: How to predict which is the game time for each player

Post by josecarlos » Thu Mar 22, 2018 8:44 pm

Crow, as always, thank you so much for your help!!! I'll read the articles that you give me and watch the video, of course!!!

Post Reply