Introduction

For years I have loved to both love and play the game of basketball. As I got older, my friends and I started to play in a fantasy league together. Each year before our draft: we would spends hours sifting through websites of predictions and rankings for the upcoming season. One day while eating breakfast I was reading through an article by ESPN about the upcoming season and a thought dawned on me; why couldn't I create my own predictions. Using Linear Regression, K-Nearest Neighbors Regressor and a Decision Tree Regressor machine learning models, I predicted the upcoming NBA season. I created this website to show my findings and give my opinion on the predicted 2022-23 season produced by these machine learning models.

Approach

My approach consisted of the following steps: collecting and storage of the data, cleaning and manipulation of the data, and finally training and testing of the data to predict the 2022-23 season.

Collecting and Storage of the Data

Before being able to do anything I had to actually get the data. I used the official API for raw NBA statistics as well as designing a web scraper using requests to pull data from NBA's advanced stats website. After pulling the data, I stored it into a MySQL database.

Cleaning and Manipulation

I spent a considerable amount of time cleaning and manipulating the data to be able to do anything meaningful with it. After the data was collected I had to determine whether each column was relevant or irrelevant. I also then had to add relevant columns to the data frame based on my own knowledge of fantasy league and online forums. My final data frame consisted of the most important player centric metrics from 2010 to 2022. A lot of sanity checking went into cleaning the data so most of what I kept and didn't keep is subjective to the person.

Training and Testing

After the cleaning of the data was complete, it was time to finally start the process of predicting the 2022-23 season. I predicted the 2022-23 season with 3 different machine learning algorithms: Linear Regression, K-Nearest Neighbors Regressor and a Decision Tree Regressor.

My dataset was split 90% to testing and the remaining 10% to testing so I would have one season outputted by each algorithm. To probably train and test each machine learning model, I created algorithm that looped through the data frame, taking a subset of the data frame in each loop where age stayed constant. In other words in each iteration, the independent variable, X = the age of the player and the dependent variable, Y = a column during the specific iteration.

Results

First I present the top 25 players predicted, using Simple Linear Regression:

SEASON_ID	PLAYER_NAME	TEAM_ABBREVIATION	AGE	NBA_FANTASY_PTS_RANK_ESPN	NET_ESPN	AVG_NET_ESPN
2022-23	Nikola Jokic	DEN	28	1	3929	47.91
2022-23	LeBron James	LAL	38	2	3685	44.94
2022-23	Kevin Durant	BKN	34	3	3433	41.87
2022-23	Giannis Antetokounmpo	MIL	28	4	3365	41.04
2022-23	Karl-Anthony Towns	MIN	27	5	3253	39.67
2022-23	Joel Embiid	PHI	29	6	3242	39.54
2022-23	Russell Westbrook	LAL	34	7	3132	38.2
2022-23	Ben Simmons	PHI	25	8	3014	36.76
2022-23	Luka Doncic	DAL	24	9	2991	36.48
2022-23	Chris Paul	PHX	38	10	2916	35.56
2022-23	James Harden	PHI	33	11	2899	35.35
2022-23	Trae Young	ATL	24	12	2877	35.09
2022-23	Pascal Siakam	TOR	29	13	2811	34.28
2022-23	Stephen Curry	GSW	35	14	2745	33.48
2022-23	Damian Lillard	POR	32	15	2718	33.15
2022-23	Anthony Davis	LAL	30	16	2683	32.72
2022-23	Julius Randle	NYK	28	17	2612	31.85
2022-23	Bradley Beal	WAS	30	18	2563	31.26
2022-23	Domantas Sabonis	SAC	27	19	2557	31.18
2022-23	Donovan Mitchell	UTA	26	20	2510	30.61
2022-23	Devin Booker	PHX	26	21	2498	30.46
2022-23	Rudy Gobert	UTA	31	22	2484	30.29
2022-23	Khris Middleton	MIL	31	22	2484	30.29
2022-23	Andre Drummond	BKN	29	24	2475	30.18
2022-23	Scottie Barnes	TOR	21	25	2458	29.98

Secondly I present the top 25 players based on the K-Nearest Neighbors Regressor Machine Learning Model:

SEASON_ID	PLAYER_NAME	TEAM_ABBREVIATION	AGE	NBA_FANTASY_PTS_RANK_ESPN	NET_ESPN	AVG_NET_ESPN
2022-23	Russell Westbrook	LAL	34	1	4655	57
2022-23	James Harden	PHI	33	2	4479	55
2022-23	Nikola Jokic	DEN	28	3	4215	51
2022-23	LeBron James	LAL	38	4	3902	48
2022-23	Stephen Curry	GSW	35	6	3752	46
2022-23	Giannis Antetokounmpo	MIL	28	6	3761	46
2022-23	Trae Young	ATL	24	7	3681	45
2022-23	Luka Doncic	DAL	24	8	3391	41
2022-23	Kevin Durant	BKN	34	9	3197	39
2022-23	Miles Bridges	CHA	25	10	3102	38
2022-23	Anthony Davis	LAL	30	10	3120	38
2022-23	De'Aaron Fox	SAC	25	12	3021	37
2022-23	Paul George	LAC	33	12	3066	37
2022-23	Hassan Whiteside	UTA	34	14	2970	36
2022-23	Andre Drummond	BKN	29	14	2942	36
2022-23	Kyrie Irving	BKN	31	17	2835	35
2022-23	DeAndre Jordan	PHI	34	17	2872	35
2022-23	Domantas Sabonis	SAC	27	17	2870	35
2022-23	Damian Lillard	POR	32	20	2820	34
2022-23	Chris Paul	PHX	38	20	2807	34
2022-23	Donovan Mitchell	UTA	26	20	2804	34
2022-23	DeMar DeRozan	CHI	33	20	2818	34
2022-23	Eric Bledsoe	POR	33	24	2676	33
2022-23	Nikola Vucevic	CHI	32	24	2672	33
2022-23	Rudy Gobert	UTA	31	24	2709	33

Finally I present the top 25 players based on the Decision Tree Regressor Machine Learning Model:

SEASON_ID	PLAYER_NAME	TEAM_ABBREVIATION	AGE	NBA_FANTASY_PTS_RANK_ESPN	NET_ESPN	AVG_NET_ESPN
2022-23	Nikola Jokic	DEN	28	1	4215	51
2022-23	Chris Paul	PHX	38	2	3993	49
2022-23	James Harden	PHI	33	4	3943	48
2022-23	Stephen Curry	GSW	35	4	3919	48
2022-23	Giannis Antetokounmpo	MIL	28	5	3459	42
2022-23	Ben Simmons	PHI	25	6	3347	41
2022-23	Russell Westbrook	LAL	34	7	3313	40
2022-23	LeBron James	LAL	38	8	3205	39
2022-23	Paul George	LAC	33	8	3186	39
2022-23	Luka Doncic	DAL	24	10	3134	38
2022-23	Anthony Davis	LAL	30	10	3120	38
2022-23	De'Aaron Fox	SAC	25	12	3021	37
2022-23	DeAndre Jordan	PHI	34	12	3073	37
2022-23	Hassan Whiteside	UTA	34	14	2970	36
2022-23	Andre Drummond	BKN	29	14	2942	36
2022-23	Trae Young	ATL	24	17	2837	35
2022-23	Kyrie Irving	BKN	31	17	2835	35
2022-23	Domantas Sabonis	SAC	27	17	2870	35
2022-23	Damian Lillard	POR	32	20	2820	34
2022-23	Donovan Mitchell	UTA	26	20	2804	34
2022-23	LaMarcus Aldridge	BKN	37	22	2739	33
2022-23	Nikola Vucevic	CHI	32	22	2672	33
2022-23	Al Horford	BOS	37	22	2702	33
2022-23	Rudy Gobert	UTA	31	22	2694	33
2022-23	Paul Millsap	PHI	38	26	2649	32

Conclusion

In my opinion the Simple Linear Regression machine learning model preformed the best. Keeping that in mind, I wouldn't base my fantasy team based on its predictions, or I would at least take it with a grain of salt. There are some intangibles and factors that I didn't take into account (the team the player plays for, the player's role, the player's injury severity history, player usage, etc). If I were to do this project again I would try to create algorithm that creates a ranking system and applies a numeric value to each of these intangibles before inputting the data into a machine learning; so the fantasy data that is predicted is more realistic. I did think for this project, that was beyond my knowledge in terms of implementing what I just proposed. At the end of the day, this project's goal was to showcase some of my skills in programming, data science and cognitive thinking. I believe my goal was met and I am satisfied with the project.

My next step is to actually create a fantasy league team based on my findings and see how it preforms.