Analysis of NBA: Moneyball Approach
Sheila Kwartemaa Boateng, Taiwo Ayeni Michael, Nahimah Suglo
2025-02-11
Introduction
In today’s world, sports like football, baseball, and basketball have become deeply integrated into entertainment and analytics. Basketball, in particular, dominates the sports landscape in the United States, with the NBA serving as a global hub for elite competition. In the ever-evolving world of professional basketball, data analytics has become a game-changer, reshaping the way teams evaluate talent, make strategic decisions, and build rosters. Much like the “Moneyball” revolution in baseball, the NBA has shifted toward data-driven decision-making, where advanced metrics influence scouting, player development, and in-game strategies.
In this project, we aim to explore key performance metrics that define great players and analyze their career trajectories. By evaluating statistical trends and applying machine learning models, we seek to uncover insights into player efficiency, career progression, and draft potential. Our goal is to bridge raw statistics with strategic decision-making, helping to identify undervalued players, predict long-term success, and provide actionable insights for team management.
Business Questions
- Identifying Undervalued Talent
Inspired by the Moneyball approach, we aim to identify players who are overlooked but possess strong potential. We focus on players who excel in scoring efficiency, playmaking, and versatility but may be underappreciated in traditional scouting. This analysis can help teams find hidden gems who could thrive in the right system.
- Predicting Player Efficiency
Using Random Forest, we predict True Shooting Percentage (TS%), a key measure of a player’s scoring efficiency that accounts for field goals, three-pointers, and free throws. By leveraging past performance data, we aim to determine which players are most efficient and what factors contribute to their success.
- Career Projection Based on Age
Player performance naturally evolves over time, but not all players decline at the same rate. Using Gradient Boosting, we analyze how a player’s career unfolds based on:
Age: A primary factor in performance trends.
Previous Season’s Stats: Recent form and consistency.
Two Seasons Ago Stats: Long-term trajectory and sustainability.
We focus on Points (PTS), Rebounds (REB), and Assists (AST), as these stats often define a player’s role and value. This model will help teams anticipate when a player may peak, plateau, or decline.
- Draft Prediction Using First-Game Metrics
Traditionally, draft projections rely on college or international performance, but in our case, we lacked pre-draft data. Instead, we analyzed players’ first professional game metrics to predict draft selection. We examined:
Scoring efficiency (PTS, FG%)
Playmaking ability (AST, TO ratio)
Defensive impact (REB, STL, BLK)
Overall productivity in limited minutes
This approach provides insights into how a player’s early performance correlates with their draft position and whether teams make the right selections based on early indicators.
Data Overview
The dataset used for this analysis is sourced from Kaggle and includes detailed statistics from various basketball leagues worldwide. The dataset comprises player performance data, including metrics such as games played (GP), minutes played (MIN), field goals made (FGM), field goals attempted (FGA), three-point field goals made (3PM), height, weight, and other relevant statistics. In total the dataset contains 53,949 rows and 34 columns.
Below is a breakdown of the key columns:
- League: Indicates which league the player is associated with (e.g., NBA, CBA, KLS).
- Season: The year of the basketball season.
- Stage: Whether the data corresponds to the regular season, playoffs, or other stages.
- Player: The player’s name.
- Team: The team the player played for.
- Performance Stats: Includes metrics such as PTS (Points), REB (Rebounds), AST (Assists), STL (Steals), BLK (Blocks), and more.
- Birth Date: The player’s birth date, which can be used to calculate age.
- Height & Weight: Physical attributes of the players.
- Draft Information: Includes draft round, draft pick, and draft team.
To ensure a comprehensive understanding of player performance and dynamics, we concentrated on players with connections to the NBA while incorporating their records from other leagues. This approach allowed us to evaluate the influence of international experience on NBA performance and the potential of undrafted players. The dataset was filtered into three main categories:
- NBA Players: All players who have participated in the NBA at any point in their careers. This group forms the core of our analysis, as the NBA’s draft system and competitive environment are central to our study.
- NBA Players Who Have Played in Other Leagues: This group includes NBA players who also competed in international leagues before or after their time in the NBA. By including these records, we aim to understand how experiences in different competitive environments impact player development and performance.
- U.S.-Born Players Who Played Exclusively in International Leagues: This category focuses on American-born players who never played in the NBA but pursued professional basketball careers overseas. Including this group helps us analyze undrafted talent and explore how these players might have performed if given an opportunity in the NBA.
Data Preparation and Cleaning
Filtering the Data
Our primary focus was to gather records of both international and local players who played in the NBA, as well as U.S.-born players who may or may not have played in the league. To ensure a structured analysis, we filtered and prepared the data by categorizing players into three main groups:
NBA Players – All players who have participated in the NBA, regardless of nationality.
NBA Players Who Have Played in Other Leagues – Players who had careers in the NBA but also competed in international leagues.
U.S.-Born Players Who Played Exclusively in International Leagues – Players born in the United States who never played in the NBA but pursued professional careers in international leagues.
# Filter NBA players
nba_players = df[df["League"] == "NBA"]
# Identify NBA players who have played in other leagues
international_players = df[df["Player"].isin(nba_players
["Player"]) & (df["League"] != "NBA")]
# Identify U.S.-born players in international leagues
us_born_international_players = df[(df["League"] != "NBA")
& (df["nationality"] == "United States")]
# Merge NBA and International Data for the same player
#(keeping all NBA rows)
merged_df = pd.concat([nba_players, international_players, us_born_international_players])
# Reset the index of the merged DataFrame
merged_df.reset_index(drop=True, inplace=True)This reduced the dataset to 21,834 rows and 34 columns.
Handling Duplicates in the Data
Duplicate records in a dataset are often indicative of data quality issues, which can arise during the data collection, import, or export processes. We identified 1,112 duplicated rows in the dataset, where the combination of player name, season, and other performance metrics were identical.
Understanding the Duplicate Issue
Duplicate rows mean that certain player-season records have been repeated in the dataset, which can distort any analysis by inflating the importance or weight of certain records. In the case of basketball data, this issue could skew performance analysis, as it might make a player appear more consistent or prolific than they actually are.
Potential Causes for Duplicates:
Data Entry Errors: Duplicates can arise when data is entered multiple times by mistake, especially during the data collection or processing phase.
System Glitches: Sometimes, technical issues during the data import or export processes (such as using different tools or systems) can lead to data duplication.
Intentional Duplicates: While it is less likely in this case, certain situations may involve repeated events, such as a player being tracked multiple times across different datasets or due to specific circumstances (e.g., tracking both regular season and playoff stats separately).
However, in the context of this analysis, it is unlikely that the duplicates are intentional, as each player’s performance is expected to appear only once per season. Having identical records could distort key analyses, particularly in terms of aggregating statistics.
Hence we removed duplicates values removed by keeping only one record for each player-season combination, ensuring that there are no multiple entries for the same player in a given season.
Handling Data Inconsistency
The Games Played (GP) column contained values greater than 82, which is higher than the typical number of games in an NBA regular season. This was likely due to the inclusion of playoff games or possible data entry errors. To explore the data, we decided to cap GP values above 82 at 82 for exploratory purposes.
The draft_round column showed a maximum value of 7, which is inconsistent with the current NBA draft format, where there are only 2 rounds. Hence we cap the draft rounds at 2
Handling Missing Values
Upon inspection, we identified several columns with missing data. The following is a summary of the columns and the number of missing values:
These missing values can arise for various reasons, such as incomplete data entry, players playing in multiple leagues, or missing player records in specific seasons. Addressing these missing values appropriately is vital for the success of our analysis.
Treatment of Missing Values
Missing values in ‘Team’ Values: The Team column had missing values for Brandon Penn and Wesley Myers, who played in the Serbian-KLS league (2019-2020). After verifying their team affiliation, we confirmed they were with KK Kolubara and filled the missing values with ‘KK’.
Missing values ‘High School’ Values: We replaced the missing high_school values with the placeholder “Unknown”. But we later drop the column entirely.
Estimating Missing ‘birth_year’ Values: The birth_year column plays an important role in understanding player trajectories and age-related performance trends. We handled missing values in the following way:
If a player had multiple records, we filled missing birth years using the most common (mode) value within their entries.
If the mode approach was unsuccessful, we used the nationality and season information to estimate the birth_year. We filled the missing values based on the most frequent birth_year for players from the same nationality and season.
For players with remaining missing birth years, we used the minimum season start year to estimate the birth_year based on the player’s earliest season.
Handling Missing ‘Height’ and ‘Weight’ Values
Height and Weight values were imputed based on the height (height_cm) of players, using the mean height or weight for players with similar physical characteristics.
If the weight was missing, we also converted it to kilograms if needed and imputed the missing values based on the mean weight of players with similar height measurements.
Additionally, the weight column, which was recorded in pounds, was converted to kilograms for consistency. Any remaining missing weight_kg values were filled using the same imputation method, based on the height_cm grouping.
Handling Missing Draft Information: For the draft_round, draft_pick, and draft_team columns, we handled missing values based on the following conditions:
For U.S.-born players and players in the NBA, missing draft information was filled with 0 for draft_round and draft_pick, and “Undrafted” for draft_team. indicating they were not selected in the draft
For international players who did not play in the NBA, missing draft information was filled with 0 for draft_round and draft_pick, and “International” for draft_team.
This ensured that all players had valid draft-related data, even if they were not drafted or played internationally.
Data Exploratory
In basketball, player performance metrics are essential for evaluating and understanding a player’s contribution to the team. These metrics provide insight into different aspects of the game, from scoring and shooting efficiency to playmaking, rebounding, and defense. Some players’ statistics are above average, while others fall below. In certain cases, the statistics are zero, which could imply a variety of factors. For example, a value of zero might indicate that the player did not participate in a particular category (e.g., no attempts in a specific stat like 3-point field goals or free throws), or it could suggest that the player did not accumulate any meaningful contributions in that stat during the season.
Shooting Efficiency
Metrics like Field Goal Percentage (FG%), 3-Point Percentage (3P%), and Free Throw Percentage (FT%) are crucial for understanding a player’s scoring ability. True Shooting Percentage (TS%) offers a more comprehensive view of a player’s shooting efficiency by considering all types of scoring attempts, including field goals, three-pointers, and free throws. High shooting efficiency is key for consistent offensive production and overall team success.
FG% (Field Goal Percentage): The average is 46.6%, which is typical for NBA players. The best shooters (75th percentile) hit over 50.6% of their shots.
3P% (3-Point Percentage): The average is 31.9%, with top shooters (75th percentile) hitting 38.9% or more from beyond the arc.
FT% (Free Throw Percentage): The average is 74.3%, with the best free throw shooters (75th percentile) making over 82.2% of their attempts.
The average is 55.7%, which is a good efficiency rate. The top 25% of players have a TS% of 59.6% or higher, indicating excellent scoring efficiency.
Playmaking & Ball Control
The Assist-to-Turnover Ratio (AST/TO) is vital for measuring a player’s ability to create scoring opportunities while maintaining possession. A higher ratio indicates effective decision-making and ball control, crucial for maintaining offensive flow and reducing costly turnovers.
The average is 1.39, meaning players generally create more assists than turnovers.
The top 25% have a ratio of 1.80 or higher, indicating excellent ball handling and decision-making.
Rebounding & Defensive Impact
Rebounds per game (REB) are an important metric for understanding a player’s contribution to maintaining possession and limiting the opposing team’s opportunities. Additionally, metrics like steals (STL) and blocks (BLK) highlight a player’s defensive impact, particularly in disrupting the opposing team’s plays and protecting the rim.
PTS_per_game: The average is 11.86 points, with the top scorers (75th percentile) averaging 14.74 or more.
REB_per_game: The average is 4.65 rebounds, with top rebounders grabbing 5.97 or more.
AST_per_game: The average is 2.29 assists, with the best passers dishing out 3.07 or more.
STL_per_game and BLK_per_game: These have lower averages (0.97 and 0.43 respectively), which is expected as these are less common events.
Performance Over Time Prominent NBA Players:
This series of line plots visualizes the performance of prominent NBA players (LeBron James, Kobe Bryant, Michael Jordan, Stephen Curry, Kevin Durant, and Giannis Antetokounmpo) over their careers, across key metrics: Points (red), Rebounds (blue), and Assists (green).
Key insights:
LeBron James shows significant improvement in points throughout his career, with notable peaks in the later years (2013-2020). His rebounds and assists also show consistent growth, indicating his all-around contribution to the team.
Kobe Bryant’s performance follows a similar trend, with sharp peaks, particularly in points. His assists and rebounds fluctuate over time but don’t show as dramatic an increase.
- Michael Jordan’s performance had a noticeable decline after 2000, especially in rebounds and assists. His points still remain relatively stable during his prime years but decrease significantly towards the end of his career.
- Stephen Curry displays impressive growth in points starting from the 2012-2013 season. Rebounds and assists increase as well, though not as drastically.
- Kevin Durant and Giannis Antetokounmpo show more steady, upward trends in their points, rebounds, and assists, highlighting their growing role as they progress in their careers.
These players demonstrate how performance evolves with age and experience, with scoring typically peaking earlier than other metrics like rebounds or assists. Notably, players who excel in True Shooting Percentage (TS%)—a key efficiency metric—continue to dominate across different stages of their careers.
Stephen Curry ranks at the top in TS% during the regular season, showcasing his elite scoring efficiency. Other standout players in this category include Kevin Durant, Shaquille O’Neal, LeBron James, and Giannis Antetokounmpo, all of whom have maintained exceptional efficiency while adapting their games over time. And they excede a composite score above 0.8, they consistently demonstrated efficiency, playmaking, and defense throughout their careers. For a detailed breakdown of these statistics, refer to the Appendix.
Players Efficiency By Age
The scatter plot, shows the relationship between a basketball player’s age and their True Shooting Percentage (TS%), a measure of scoring efficiency. Players in their early careers, typically between 17 and 23 years old, exhibit greater variability in TS%, likely reflecting skill development and adaptation to the NBA. During their prime years, from approximately 24 to 32, players tend to cluster more densely within a TS% range of roughly 50% to 65%, suggesting a stabilization of performance. As players progress into their later careers, from ages 33 to 45, there appears to be a slight decrease in TS% for some, coupled with increasing variability, potentially due to physical decline or evolving roles within their teams.
Change in Stats
Some NBA players are international talents who transition to the NBA, while others continue their careers in international leagues after leaving the NBA. This shift in playing environments can lead to significant changes in player statistics.
Average Change in Stats - International to NBA Transition:
PTS_change 126.153846
REB_change 46.692308
AST_change 34.769231
STL_change 5.923077
BLK_change 4.461538
MIN_change 226.992308
dtype: float64
******************************************************************************************
Average Change in Stats - NBA to International Transition:
PTS_change 89.800
REB_change 26.900
AST_change 26.550
STL_change 7.250
BLK_change 4.650
MIN_change 113.615International to NBA Transition:
For players who moved from other country to play in the NBA, the result shown that;
- PTS ChangePlayers transitioning from international leagues to the NBA show a large spread in points change, with some players experiencing a significant increase in points.
REB Change: The change in rebounds is much less variable than points, though some players do experience a drop or moderate increase.
STL, BLK, and MIN: There is a noticeable decline in steals and blocks upon entering the NBA, possibly due to the more physical and competitive environment. Minutes played (MIN) also drops in the transition, which could be linked to the role players take on in the NBA.
NBA to International Transition:
PTS Change: Players who move from the NBA to international leagues generally experience a decline in points, although there are outliers who maintain or improve their performance.
REBs and AST: Like points, rebounds and assists also decrease in many cases.
STL, BLK, MIN: These metrics show a slight increase for some players transitioning to international leagues, which could reflect the more prominent role these players take on in their teams abroad.
These findings indicate that international players generally see significant improvements when transitioning to the NBA, especially in scoring, rebounding, and assists. The NBA’s pace, coaching, and playing conditions may contribute to these improvements, allowing international players to elevate their overall performance. However, the magnitude of change might vary depending on a player’s role, adaptation period, and playing style.
On a broader context: International players have reshaped the NBA with diverse playing styles, emphasizing teamwork, ball movement, and versatility. Stars like Nikola Jokić, Giannis Antetokounmpo, and Luka Dončić have demonstrated that international players can not only adapt but dominate in the NBA. The data supports the hypothesis that international players often improve their performance when transitioning to the NBA https://www.basketballnews.com/stories/an-inside-look-at-how-international-players-are-taking-over-the-nba
Drafted Vs Undrafted
In the NBA, drafted players are those selected through the league’s annual draft process, typically receiving guaranteed contracts and structured development opportunities. Undrafted players, on the other hand, enter the league without a draft selection, often having to prove themselves through training camps, summer leagues, or international play.
Drafted players generally outperform undrafted players across key metrics, including points (PTS), rebounds (REB), assists (AST), steals (STL), and blocks (BLK). Specifically, drafted players record a median of 399 points, 156 rebounds, 63 assists, 30 steals, and 12 blocks per game, whereas undrafted players have a median of 323 points, 112 rebounds, 48 assists, 26 steals, and 6 blocks.
The differences in performance can largely be attributed to opportunities and role expectations. Drafted players typically have a higher chance of securing significant playing time, receiving more structured development, and being positioned for key roles within their teams. Conversely, undrafted players often have to fight for roster spots and may receive fewer minutes, affecting their statistical output.
Undervaluing Players: A Moneyball Approach
Given the metrics above, we wanted to identify players who may be undervalued in the NBA, especially those whose performance metrics suggest they could contribute more if given the opportunity. This aligns with the Moneyball approach, where teams seek players who are overlooked by conventional evaluations but excel in key statistical areas.
Undervalued Scorers
Players like D.J. Richardson (0.345 PTS/MIN) and Mike Myers (0.324 PTS/MIN) exhibit exceptional scoring efficiency with limited minutes, making them potentially valuable assets if utilized more. Richardson, for example, scored 21.5 points in just 62.2 minutes, suggesting he could have a much larger impact with more court time. While players like Jimmer Fredette and Jeff Withey show strong scoring, their lower efficiency per minute suggests they were less effective despite potential for greater contributions. From a Moneyball perspective, these players would be considered undervalued because their efficiency doesn’t match their recognition, and with the right opportunities, they could have significantly impacted teams looking for efficient scoring.
Good Passers
Pass-first players like John Stockton and Chris Paul are well-known for their elite playmaking abilities, but lesser-known players like Kirk Hinrich and Patty Mills also demonstrate high assist-to-turnover ratios and solid assists per game. These players may not have the spotlight but play an essential role in ball distribution, offering depth in playmaking, especially for teams in need of efficient facilitators. Their contributions are often overlooked, yet they provide significant value to teams by maintaining ball control and ensuring smooth offensive execution.
Versatile Players
Players like Vince Carter, Gary Payton, and Tim Duncan were often undervalued for their all-around contributions. Despite their ability to impact both ends of the floor, their multi-faceted skills were sometimes overshadowed by flashier stars. Teams could have benefited greatly from these players’ versatility, especially in terms of leadership, defense, and overall contribution.
Three-Point Specialists
With the growing importance of three-point shooting in modern basketball, players like Marcus Haislip and Jimmer Fredette stand out. Haislip’s 65% 3P% and Fredette’s 60.3% show their potential as elite shooters, though their careers didn’t reflect their abilities due to limited opportunities. Similarly, Peyton Siva and James Nunnally, despite their strong three-point shooting overseas, weren’t fully utilized in the NBA. In today’s game, these players could have stretched the floor and created more scoring opportunities, making them valuable contributors for teams looking for reliable shooters.
Modeling
Efficiency Prediction (TS%)
True Shooting Percentage (TS%) is a key metric in evaluating a player’s scoring efficiency, accounting for field goals, three-pointers, and free throws. As observed above, many of the most prominent players maintained a high TS%, highlighting its importance in measuring offensive effectiveness.
To build a predictive model for efficiency (TS%), I selected specific statistics based on their relevance to player performance and their availability in the dataset:
PTS, REB, AST, STL, BLK, TOV: Core box score statistics that contribute directly to a player’s overall efficiency.
FG% and FT%: Essential for calculating a player’s shooting efficiency.
MIN: Playing time determines opportunities to accumulate stats and influence the game.
Age: Experience and physical prime can impact a player’s efficiency.
The data was split into 80% for training and 20% for testing, ensuring a balanced model evaluation.
Model Evaluation
Random Forest was applied to predict TS%. The performance of the model was assessed using the Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared (R²), and Adjusted R-squared (Adjusted R²). The following observations were made:
Before parameter tuning:
R² = 0.7659, meaning the model explained 77% of the variance in player efficiency (TS%).
RMSE = 0.03
MAE = 0.0226
After parameter tuning:
R² = 0.7671, a slight improvement but still approximately 77%.
MAE decreased by 0.001, showing minor improvement.
RMSE remained the same (0.030).
TS% Prediction Results:
Best Parameters: {'max_depth': None, 'min_samples_leaf': 2, 'min_samples_split': 2, 'n_estimators': 200}
Root Mean Squared Error: 0.03010722646594936
Mean Absolute Error: 0.022501884921930802
R-squared Score: 0.7671658127820751
Adjusted R-squared Score: 0.7656135848672889
Feature Importance:
feature importance
6 FG% 0.684835
7 FT% 0.139206
1 REB 0.030102
0 PTS 0.026531
5 TOV 0.022990
2 AST 0.021842
8 MIN 0.020237
9 age 0.019064
4 BLK 0.018954
3 STL 0.016241Key Feature in TS% Prediction
- Field Goal Percentage (FG%) had the highest influence, contributing 67.3% to the model’s predictive power.
In general we can say that the tuned model was best, although the R-square was still around 77%.
Career Trajectory Prediction
As observed, player performance tends to decline with age, though exceptions exist LeBron James, for example, continues to play at a high level even past the typical prime years. To analyze NBA career trajectories, we applied a Multi-output Regression model using Gradient Boosting, predicting three key stats: Points (PTS), Rebounds (REB), Assists (AST)
Model Performance After Tuning (Best Model)
| Fundamental Stats | R² | RMSE | MAE |
|---|---|---|---|
| PTS (Points) | 0.4448 | 404.71 | 298.53 |
| REB (Rebounds) | 0.3999 | 161.80 | 118.09 |
| AST (Assists) | 0.5478 | 105.16 | 70.24 |
Feature Importance Analysis
prev2_PTS (two-season lagged points) is the most critical variable in predicting future PTS.
prev_PTS (last season’s points) and prev_REB (last season’s rebounds) also contribute significantly.
Age and assist-related metrics (e.g., prev2_AST) have less influence on projections.
Tuning the model did not significantly change the ranking of important features.
Key Insights
Impact of Performance Metrics
Recent seasons’ performance (prev2_PTS, prev_PTS, and prev_REB) strongly predicts a player’s future trajectory, particularly in scoring.
Assists have the highest predictability (R² = 0.547), while rebounds have the lowest (R² = 0.399), suggesting different patterns of skill development.
Broader Career Implications
In a broader sense Playmaking skills (AST) develop more consistently, while scoring and rebounding are more volatile. Scoring and rebounding trajectories are harder to predict, likely due to external factors like coaching strategies, role changes, and injuries. Moderate predictive power (R² ~ 0.40 - 0.55) suggests that while general trends exist, individual career paths can vary widely.
While NBA career trajectories can be modeled with moderate success, they remain highly individualized. This highlights the complex nature of player development, emphasizing the need for flexible and personalized training, scouting, and career planning strategies in professional basketball.
Drafted Status Prediction (SVC)
The goal of this analysis is to predict whether a player will be drafted into the NBA based on their performance metrics from their first season. These metrics serve as a proxy for their overall abilities and potential, as we did not have access to their college performance data. The model classifies players as either drafted (Class 1) or undrafted (Class 0) based on their early performance in the league.
Before hyperparameter tuning, the SVM model achieved 79.55% accuracy, with 679 undrafted players correctly classified and 91 drafted players identified. However, 115 drafted players were misclassified, leading to a 44% recall for drafted players and a precision of 52%.
SVM Classifier:
Accuracy (Balanced): 0.7954545454545454
Confusion Matrix (Balanced):
[[679 83]
[115 91]]
Classification Report (Balanced):
precision recall f1-score support
0 0.86 0.89 0.87 762
1 0.52 0.44 0.48 206
accuracy 0.80 968
macro avg 0.69 0.67 0.68 968
weighted avg 0.78 0.80 0.79 968After tuning, the SVM model achieved 83.06% accuracy with optimized parameters (C = 10, gamma = 1, kernel = ‘rbf’), balancing true positives (TP) and true negatives (TN) while minimizing false predictions. The confusion matrix shows 669 undrafted players correctly classified and 135 drafted players identified, with 71 drafted players misclassified as undrafted.
The goal was to improve recall for drafted players (66%) while maintaining precision (59%), ensuring a fair balance between identifying true drafted players and avoiding false selections. This trade-off reduces misclassifications while keeping a strong overall performance.
Best Parameters from Grid Search:
{'C': 10, 'gamma': 1, 'kernel': 'rbf'}
SVM Classifier:
Accuracy (Balanced): 0.8305785123966942
Confusion Matrix (Balanced):
[[669 93]
[ 71 135]]
Classification Report (Balanced):
precision recall f1-score support
0 0.90 0.88 0.89 762
1 0.59 0.66 0.62 206
accuracy 0.83 968
macro avg 0.75 0.77 0.76 968
weighted avg 0.84 0.83 0.83 968Conclusion
In the ever-evolving landscape of professional basketball, where advanced metrics increasingly influence team strategy, our study aimed to bridge the gap between raw statistics and strategic decision-making. Inspired by Moneyball’s approach, we sought to unearth hidden talent, project player efficiency, and anticipate career trajectories, enriching traditional scouting with data-driven insights. Our analyses, spanning from True Shooting Percentage prediction via Random Forest to career projection using Gradient Boosting, highlighted both the promise and the complexity of quantifying player performance. Visualizations such as the ‘Player Efficiency (TS%) by Age’ scatter plot further underscored the dynamic nature of career progression, with youth holding potential, the prime showcasing consistency, and later years demanding adaptability.
Recommendation
Embrace Multifaceted Evaluation: Rather than relying solely on traditional scouting or advanced metrics,teams should synergize these approaches to gain a more comprehensive understanding of player potential, performance, and fit within the team.
Invest in Talent Development and Coaching: Acknowledge the uniqueness of each player’s career trajectory by tailoring development plans to address specific areas for improvement,considering age and role evolution.
Strategically Leverage Analytics in Scouting: Utilize advanced metrics like True Shooting Percentage to identify undervalued players, but consider the potential and the specific team context in which the player will be utilized. This is especially true with age and projected output, for planning beyond one season.
Cautious Use of First-Game Metrics: The first impression and performance should not be weighted significantly due to its volatility.
Ultimately, as basketball continues to evolve, so too must the strategies for talent acquisition and team building. By embracing a holistic approach that marries data analytics with strategic decision-making, teams can optimize their performance and unlock unparalleled success. Through continuous adaptation and a commitment to innovation, basketball organizations can harness the full potential of their players and teams.
Reference
Alperendes. (2021, March 3). NBA Analysis - How Basketball has changed? Kaggle. https://www.kaggle.com/code/alperendes/nba-analysis-how-basketball-has-changed
Lewis, M. (2003). Moneyball: The art of winning an unfair game. W. W. Norton & Company.
Kubatko, J., Oliver, D., Pelton, K., & Rosenbaum, M. (2009). Basketball on paper: Rules and tools for analytical decision-making. Potomac Books.
Skansi, S. (2017). Basketball data science: With applications in R. CRC Press.
Fewell, B. M., Armatas, V., Neville, J., McGuire, R.T., Williams, M. R., & Davids, K. (2012). Modeling representative task constraints in Australian football. Journal of Science and Medicine in Sport, 15(2), 153-159.
Appendix
Top 10 Efficient Scorers:
Player Team TS% PTS_per_game MIN Efficiency
8012 Primoz Brezec KRY 0.855696 20.833333 183.8 0.113348
8855 Jimmer Fredette PAN 0.824176 13.666667 342.3 0.039926
16474 D.J. Richardson SPI 0.820611 21.500000 62.2 0.345659
20199 Jeff Withey IRO 0.811924 14.888889 246.7 0.060352
16442 Seth Tuttle LIM 0.782584 18.333333 178.5 0.102708
15525 Tony Gugino RIL 0.774246 12.666667 182.6 0.069368
3006 Raja Bell PHX 0.772025 13.600000 214.8 0.063315
16992 James Nunnally FEN 0.771899 11.866667 635.1 0.018685
17516 Cory Bradford BOS 0.770021 15.000000 92.4 0.162338
17544 Mike Myers OOS 0.767045 13.500000 41.7 0.323741
Top 10 Good Passers:
Player Team AST_per_game AST_TO_Ratio
3708 Kirk Hinrich CHI 4.000000 10.000000
7186 Patty Mills SAS 3.571429 8.333333
7582 Joe Ingles UTA 4.714286 8.250000
17534 Maarty Leunen AVE 5.125000 8.200000
611 John Stockton UTA 11.400000 8.142857
5874 Zaza Pachulia DAL 3.200000 8.000000
2021 Earl Watson MEM 3.750000 7.500000
5811 Chris Paul LAC 7.250000 7.250000
19931 Cliff Clinkscales HAL 10.125000 7.147059
13113 Matthew Pettit MIN 3.500000 7.000000
Top 10 Versatile Players:
Player Team PTS_per_game REB_per_game AST_per_game \
1 Vince Carter TOR 25.695122 5.804878 3.926829
2 Karl Malone UTA 25.548780 9.500000 3.707317
4 Gary Payton SEA 24.170732 6.451220 8.926829
6 Grant Hill DET 25.756757 6.621622 5.202703
7 Kevin Garnett MIN 22.925926 11.802469 4.950617
8 Michael Finley DAL 22.621951 6.317073 5.341463
9 Chris Webber SAC 24.453333 10.493333 4.600000
10 Ray Allen MIL 22.060976 4.378049 3.756098
12 Tim Duncan SAS 23.189189 12.405405 3.162162
13 Glenn Robinson MIL 20.901235 5.987654 2.382716
STL_per_game
1 1.341463
2 0.963415
4 1.865854
6 1.391892
7 1.481481
8 1.329268
9 1.600000
10 1.341463
12 0.891892
13 0.962963
Top 10 Three-Point Specialists:
Player Team 3P% 3PA
12656 Marcus Haislip DON 0.650000 160
8855 Jimmer Fredette PAN 0.602564 78
8853 Paul Zipser BAY 0.579710 69
8783 Paul Zipser BAY 0.579710 69
16992 James Nunnally FEN 0.571429 119
15328 Chasson Randle NYM 0.548077 104
20438 Peyton Siva BER 0.541667 72
19714 Peyton Siva BER 0.541667 72
6021 Pau Gasol SAS 0.538462 104
16387 Jon Diebler GAL 0.538462 91Top 15 Players - Regular Season TS%: Player Stephen Curry 0.623333 Kevin Durant 0.612656 James Harden 0.609455 Kawhi Leonard 0.594411 Shaquille O'Neal 0.589213 LeBron James 0.587365 Chris Paul 0.585591 Giannis Antetokounmpo 0.584268 Dirk Nowitzki 0.580947 Kobe Bryant 0.555138 Dwyane Wade 0.551497 Tim Duncan 0.548705 Allen Iverson 0.523202 Michael Jordan 0.491147 Name: TS%, dtype: float64 Top 15 Players - Playoffs TS%: Player Kawhi Leonard 0.611560 Stephen Curry 0.605011 Kevin Durant 0.594372 LeBron James 0.585926 Chris Paul 0.583343 James Harden 0.581835 Dirk Nowitzki 0.569571 Giannis Antetokounmpo 0.557766 Shaquille O'Neal 0.552266 Kobe Bryant 0.544648 Tim Duncan 0.543229 Dwyane Wade 0.541905 Allen Iverson 0.496732