How to approach Scouting Reports for players?

For those of you that don't know me or my work, I've been developing a basketball statistical analysis tool for some time now. You can find out more about it here. This tool is also my thesis, and for whatever reason, I thought it'd be more impressive if I put "Natural Language capabilities" in the thesis' title, so now I have to do exactly that. What the goal of those capabilities are is to offer a quick scouting report on a team or a player if you can't bother to go over all the tables and numbers and stats, and just want a quick overview.

I've already implemented a basic version of what I'm trying to do for teams, like so:


The way the teams' Scouting Reports work is quite naive right now, but should be sufficient to persuade my professors that I've done something natural language-like so that I can graduate this November. It checks the team's ranking at each stat, and considers whether that team's any good at that based on whether it's on the Top 5 teams, Top 10, 10-20, or 20-30 for that stat. I know that a team's ranking isn't sufficient, but it's the best I could come up with for a start back when I first implemented this. I'll change the ranking thresholds so that they're not hard-coded to 5, 10, 20 and 30, since the tool should be compatible with any league, not just 30-team leagues.

Now I want to do the same for Players, but I wanted to ask you guys for your suggestions on how I should approach this. Given all the information that NBA Stats Tracker calculates and provides, what should the Scouting Report consist of? An analysis on his best 4 and worst 4 averages? A full analysis of every aspect of his stat-line? Which factors should I be using to consider whether the player's very good, good, decent or plain bad on each stat? Are basic stats and averages enough, or should I be using metrics (a lot of them are already implemented)? What should I be using as thresholds? Ranking percentages (e.g. whether a player is on the Top 10% of the league in FG% or PER or TS% etc.)? What differences should I look out for per position?

Just throwing ideas around, but it seems the metric stat line usually paints a good picture of a player. You could probably take the basic route and talk about that players past performance about the team they're facing, along with past games they just recently played, then include how they're doing overall through the season. You could have an advanced tab that covers the metric stats covering the same stuff. As you probably know there's a lot of ways to attack it.

Like you said their best and worst averages is good enough but that I think is a given. For me since your doing a scouting report for a player maybe include their tendencies."Player likes to shoot 3 points/attack the basket/commits foul" Something like that. It would look like a true report.

I personally would use various statlines. PER, Winshare, ORTG, DRTG, usage, Preferred shooting spot/distance, ETC. So many stats you can use, but I don't know how your program will retrieve data/database automatically unless you plan to hard code them for each players.

Interesting school project you're doing. This looks like Visual Studio with C# job. Right?

Personally, I'd like to see a program that ranks players for the use of Fantasy League. Probably with features like, a)that tells how many games a player will play in upcoming week with a click of button, b)rates player's performance in previous week or two, c)shows where a player ranks in his team's depth chart positionally, ETC.

Yes, it is done in C#, uses WPF and .NET 4.

To help you help me, this tool only contains stats you see in a box score right now, although I'm planning to have much more advanced stats (such as shot percentages per position on the floor, etc.) after the November graduation. Right now, just things you can calculate by accumulating box scores, so basic stats, averages, and metrics. No tendencies.

The tool works with whatever data you give it. You can either download NBA data from B-R automatically, you can import data from your NBA 2K save, or create a database with teams and players from scratch to use in whatever league you want, and then accumulate stats by inputting box scores, or by manually editing the accumulated stat lines of teams and players.

Thanks for all the suggestions so far, keep them coming. I should start working on player Scouting Reports when I finish up the performance graphs for players and teams.

I am surprised your school lets you do your own thing as a school project. Should be a lot of fun. I'd probably have done something similar to yours if my school offered the same, but instead I worked with some stupid companies making them photo storage app and some esl/real estate web pages.

Great work btw. Perhaps I will download your program and give it a try to get a better feel before I give you anymore of my lame-ass suggestions.

Public education baby.

And in any case, it's my thesis. The way this works here is that each professor gets a catalog of subjects out, and each student applies for the subjects they like the most out of all the announced. If no subject is to their liking, they can find a professor that is close to the field of what they want to do, and if the professor likes that particular idea, they can have their own subject become their thesis. So we're allowed to be creative with our thesis' subject, as long as it's something innovative and makes use of the latest technologies, or improves upon them.

Yeah, please do download NST and try it out. Maybe after you get accustomed to all the information and analysis it offers, you may come up with some better suggestions on how to make the Scouting Report up, based on information that's already there or that is easy to implement.

I realize your points about more advanced stats than the ones already implemented, but I'm currently looking to meet the mid-October deadlines for the November graduation, so I don't want to go too deep in implementing new stuff. Just a basic, helpful summary of what the user needs to know for a particular player without going through all the information.

I'll be coming back for suggestions on how to make NBA Stats Tracker even better after November/December as I'm going to continue developing it to make it as much of a full-fledged statistical analysis tool as possible, but until then I just want to complete the natural language capabilities and graduate.

EDIT: As a basic starting point, what top percent of players would you consider good at each stat, or should I be using something like (max - (max - min) * percent) or (max - (max - avg) * percent) as a bottom threshold?

For example, if the maximum RPG for the Centers of a league is 13 and the average is 6, with a 30 percent factor the second formula would become (13 - (13 - 6) * 0.3) = 13 - 2.1 = 10.9 so only players between 10.9 to 13 RPG would be considered great rebounders. I could then use something like 75 percent for good rebounders, and so on. What do you think?

EDIT 2: Obviously the "Rebounds per 36 minutes" metric would be much more suitable here, and I have that one available, so I'd probably use that in the above example instead of "Rebounds per game".

Personally, I'd like to see a program that ranks players for the use of Fantasy League. Probably with features like, a)that tells how many games a player will play in upcoming week with a click of button, b)rates player's performance in previous week or two, c)shows where a player ranks in his team's depth chart positionally, ETC.

That still will not help you win.
The first two that you want are already in Yahoo Fantasy Leagues with the click of a button.

Which factors should I be using to consider whether the player's very good, good, decent or plain bad on each stat?

What should I be using as thresholds?

Start with the league average for the particular season you're basing on and go on from there to determine and compare good and bad players in that particular season.
You could also do historical comparison of stats but I guess that would be unnecessary considering your aim for now is a 'simple' program.

Are basic stats and averages enough, or should I be using metrics (a lot of them are already implemented)?

Depends on your target demographic that will 'use' your program. If you're going for casual fans why bother with metrics. That advanced tab for metrics might be good alternative if your aim for the program is for easy to read and casual approach to a scouting report.

I'm trying to hit a good tradeoff of being on the way to making this a tool not just for casual fans, but for coaches, etc, but also something I can do to meet my deadlines.

I could do something "basic" for now and really put the whole shenanigan out after November. But I don't want to stop at something "simple", I want this to become something coaches and hardcore fans will want.

So what I'm asking is pretty much both what I should do for a "casual" approach now, and what I should do to really create some good scouting reports based on everything the tool keeps track of later on. Oh, and what I should add to the tool to keep track of in order to get it to be as advanced and detailed as possible.

EDIT 2: Obviously the "Rebounds per 36 minutes" metric would be much more suitable here, and I have that one available, so I'd probably use that in the above example instead of "Rebounds per game".

Rebounding %. Since you have the team numbers.

And there's a historical upper "limit" to it.