nba-expected-bucket

Predicting the outcome of shots based on the events and tracking data available for the 2015/16 season.

Quick links:

Data

The repo combines events data and tracking data from the 2015/16 season.

Data Source:

Events data: NBA events and shots api
Tracking data: https://github.com/linouk23/NBA-Player-Movements

Features used in the model

Categorical Features

SHOT_TYPE	SHOT_ZONE_BASIC	SHOT_ZONE_AREA	SHOT_ZONE_RANGE	ACTION_TYPE_SIMPLIFIED
2PT	Mid-Range	Right Side Center(RC)	16-24 ft.	Layup
3 PT	Above the Break 3	Right Side(R)	24+ ft.	Dunk
	Restricted Area	Center(C)	8-16 ft.	Hook Shot
	In The Paint (Non-RA)	Left Side(L)	Less Than 8 ft.	Jump Shot
	Left Corner 3	Left Side Center(LC)
	Right Corner 3

Numerical Features

AVERAGE_DEFENDER_DISTANCE	SHOOTER_DEFENDER_DISTANCE	OFFENSE_SPACING	DEFENSE_SPACING	SHOT_CLOCK

Calculating Spacing

I used convex hull technique to calculate the OFFENSE_SPACING and DEFENSE_SPACING. This is explained much better here: http://projects.rajivshah.com/sportvu/Chull_NBA_SportVu.html

AVERAGE_DEFENDER_DISTANCE and SHOOTER_DEFENDER_DISTANCE are calculated using the distance formula.

Where do teams take their shots from?

Does better spacing result in higher shooting percentage?

From the charts below we can infer that, although that is not the case always, we do see that among the teams with shooting percentage higher than the league average, the spacing also is better and vice versa