You, along with the other 3.
Nylon Questions: How can we predict performance for incoming NBA players?
Kidd continued to improve as a player through his mid-to-late 20s, while Thomas had a long peak and led the Detroit Pistons to two championships. So both are favorable comps for Wall. Francis and Anderson are less favorable.
It has fewer bells and whistles. It starts with some basic biographical information for each player. NBA players, like MLB players, improve on average through about age 27 and then begin to decline after that. Players chosen with an earlier draft pick tend to have a higher ceiling, meanwhile, even once we control for other variables. We start with a few statistics related to his scoring and shooting ability. For more precise definitions of these, see Basketball-Reference.
Players like LeBron James and James Hardenwho rank highly in both usage rate and true shooting percentage, are the best scorers in the game, providing both volume and efficiency. Although less important to his overall value, it provides a purer gauge of shooting ability than true shooting percentage, which reflects both shooting ability and shot selection. Three-pointers, meanwhile, remain more efficient than 2-pointers, on average.
His rebound rate is the share of rebounds he grabs while on the floor 10 percent is average. These statistics can sometimes tell a reasonably complete story about each player. On the downside, Wall commits a lot of turnovers. And he neither shoots all that many threes nor draws all that many fouls, which can make his game flat at times.
Historical players start with a perfect similarity score ofand points are subtracted for every difference. The process sounds complicated, but the comparisons are sometimes intuitively satisfying. As a Pistons fan growing up, for example, I can see the similarities between Wall and his No.
Compare their stats on Basketball-Reference. Even so, the comparison is not perfect. Thomas drew more contact around the basket, resulting in more free-throw attempts. Like snowflakesin other words, no two NBA players are exactly alike.Our second project at Metis was called Project Luther and it was about web scraping and linear regression.
After several hours of brainstorming, I decided to try create a model that would predict the winning percentage of a given NBA team at the start of the season before any games have been played. My goal was to try to take this model to see if one could potentially use it to make money by gambling. I decided to pull my statistics and information from the website basketball-reference.
This website has an enormous amount of statistics for each season going back decades. I ended up pulling about 40 years worth of various stats, including overall team stats, coach information, and player information. I used BeautifulSoup in order to scrape this information from the website. Once I had the data in a single pandas dataframe, I immediately separated out a portion of my data to be used for final testing. I chose to holdout the season, since I wanted to combine my predictions for that year with gambling data.
I then ran a baseline linear model to get a starting point. The feature I chose was average margin of victory from the previous season because I felt that would be a decent predictor of the success of the following season. My next step was to try running all my features together. After narrowing down my feature list, I tried adding polynomial terms to see if that could improve my model. Throughout my feature engineering, I used Lasso cross validation to test my model, and to continue removing unneeded features.
In the end, my final model included Lasso regularization and included features such as average age of the players, margin of victory, number of points scored, number of returning players, and number of blocks.
The root mean squared error looks great at first glance, until you remember that I was predicting a percentage. That RMSE is equivalent to about 9 games of an 82 game season which is pretty variable. My next question was can this model be used to win money by gambling. Vegas gives each team an expected number of wins in the upcoming season and you then bet that the team will either win more games over or win less games under.
If you end up being correct, you win. This seems great but I would want to test this on other years before believing that number was something other than happenstance. The residual plot below shows that my model is over-predicting when the actual winning percentage is low and under-predicting when the actual winning percentage is high.
I picked a few examples of where my model prediction was very off and found that there were consistent reasons why. The following table shows the team, year, the difference between my prediction and the actual winning percentage, and the reason why my prediction was so wrong. I believe if I incorporated that information, my model would be a much better predictor. Sign in. Anders Olson-Swanson Follow.Blinders Blinders.
About the Writer: After much research and preparation, Blinders turned his vision and passion for fantasy sports into the first Daily Fantasy sports site with Salary Cap based games in June of FantasySportsLive. He is a longtime online poker player and blogger, and the only daily fantasy grinder who was willing to take on Buffalo66 in his multi-sport fantasy challenge.
Baseline fantasy strength is the average fantasy score your player would make against an average defense, over an infinite number of trials. The match-up factors account for all of the game specific factors mostly knowable that will tend to make your player score higher or lower than their baseline strength. The game variance, accounts for all of the unknown things that can and do happen in actual games.
Assuming that players are priced at or near their baseline strength like we do at FSLthe skill in daily fantasy sports is applied in the match-up factor area. Nothing can really be done about game variance, other than having or hoping you have a large enough bankroll to get to the long-term.
Blinders: Projecting NBA Players
Long-term in daily fantasy sports is in the thousands of leagues range. I wanted to go into some depth on how to determine the defensive match-up factor for fantasy basketball. The defensive match-up factor is a primary component of the game match-up factors, but does not include other factors like the injury status of your player, the hotness of your player, if your player is expected to get more or less minutes than normal, if the game is at home or away, and other game specific factors.
To get to the defensive match-up factor you really need to go to the detail of projecting individual player statistics in fantasy basketball. The problem with fantasy basketball is that you have 5 unique positions, but team defensive statistics are not broken down by type rushing, passing, receiving like is done with football.
There are a few sites that will give you a defensive strength grid by G, F, and C but that is really not going to be very accurate at the individual player level. There are just too many exceptions in the NBA. You have pure point guards and high scoring point guards. You have defensive and offensive specialists. You have players that are good rebounders but poor scores. You have players who could be listed and play at multiple positions. The list goes on and on, and a simple defensive strength against a plain Vanilla Guard is just not going to cut it.
So to see how your player is expected to perform, you need to understand the defensive strength factors for each and every statistical category that is scored at the site you are playing at. To get the baseline strength number you multiply the averages at each category points, rebounds, assists, steals, blocks, turn-overs by the fantasy points received at each category and sum the results.Sports analytics is a constantly evolving field and keeping up can be a challenge, especially with so much work being divided between the public and private spheres.
Our staff compiled 10 questions whose answering will likely guide the next few years of public analytic work. Hopefully, these questions will help spark, refocus, and recalibrate conversations and lead to collaborative progress here at Nylon and everywhere else sports analytic work is being done.
Predicting future performance in any occupation is hard, especially in professional sports. Evaluating and selecting the best players is the quickest route to success. In the NBA, player selection is more important than in other professional sports leagues because of the way the game is structured.
First, basketball is played with only five people on the court which is the fewest of any major professional sport.
Baseball has nine, football and soccer have 11, and hockey plays with six. With fewer players on the court, the best players have more of an impact in basketball compared to other sports. Two All-Stars on a basketball team constitutes an average of about 14 percent of an NBA roster and 40 percent of the players on the court at a time.
In the NFL, two All-Pros only makes up about four percent of the roster and 18 percent of the players on the field at a time. Picking better players is more important in the NBA because it has more of an impact. A second reason for the importance of stars is that basketball players are required to play both offense and defense. Most sports have specified positions for offense and defense, the most obvious is football.
Unless your name is Deion Sanders, Bo Jackson, or Charles Woodson, football players rarely play both sides of the ball. Even soccer and hockey have specified offensive and defensive players.
In basketball, every player has to contribute on both ends of the court. If talent is so important what can we do about improving our ability to predict the performance of incoming NBA players?
With the growing popularity of analytics in sports and data science in many business operations, our ability to predict performance should only increase. More and more people are realizing the potential of machine learning and working to understand how to implement algorithms to help solve problems. But analytics alone cannot solve this problem, the interweaving of analytics with traditional scouting practices will lead to better results. The basic premise is to allow complicated algorithms to do most of the work and let humans sort out the rest.
Imagine a scenario where an algorithm can take all the data provided in pre-draft tests and performances to select the top prospects. From there NBA scouts and front office personnel can perform their own due diligence by watching film, scouting games, and interviewing players. Combining machine learning with typical scouting efforts can improve efficiency and ability to gain more insights into incoming players.
Using math, machines, and human knowledge to select players is bound to improve the process, but we must emphasize the correct data. To help understand areas of improvement in data collection we can use ideas from Fergus Connolly, a leading sport and human performance expert.
The four coactives of physical, psychological, technical, and tactical skills act in unison during an athletic performance and together they influence the performance of teams and players.
The physical skills include height, length, strength, and leaping ability. Tactical skills, also known as basketball I. In most sports, and the NBA is no exception, we focus most of our attention on the physical attributes.
Think about it, the NBA combine is a multiple day event promoting the physical attributes of incoming players. Finally, the players play in a multitude of pick up style games against their fellow prospects. One way we could improve the prediction of incoming players is to gather more data from all four coactives. We focus on physical skills in abundance followed closely by technical skills for prospects. On the other hand, tactical skills are a vital part of evaluating players but we do not have an easy way to quantify it from an analytics point of view.
If there were a way to quantify the tactical skills of a player and how they benefit their team it would go a long way. Tactical skills such as movement without the ball or help defense and how it benefits their team are important parts of the game.Staff Members. Future Draft Picks. NBA Draft History. Salary Cap. This summer I hope to make my predictions methodology more transparent to the reader, and as always I am looking for ways to improve the model.
Data are from seasons, all D1 teams. Of course just looking at the raw numbers can be a bit misleading because of players transferring out. Typically extremely inefficient players either transfer to smaller programs or stop playing D1 basketball completely. But if we limit the data to 4-year players, the pattern is still there:.
Predicting NBA Winning Percentage
Returning freshmen typically show the greatest improvement. Keep in mind that this is the average and that individual development patterns can vary widely. I will try discuss the unpredictability of player development in a future column, but the main point I want to emphasize is that this is not simply a case where everyone gets a little better every year.
What happens is some players show dramatic improvement, some players tread water, and a few players get worse. Not surprisingly, this also shows up in the team data. Teams that return more freshmen minutes are more likely to improve the following season. To my knowledge, my preseason predictions model is the only model that incorporates the importance of freshmen development.
When we predict the standings in a conference, we like known commodities. None of those players was particularly dominant and that means that a lot of people are going to write West Virginia off next season. Different Development Curves? One of the things I have been thinking about lately is whether we need to think more about the development curves for different types of players. Last summer Drew Cannon wrote about how big men develop more slowly than guards.
But last week when I was writing a look-back column on Frank MartinI was thinking how we should really break out other types of players too. In particular, three point shooters are significantly more likely to be efficient off-the-bat, and significantly more likely to be efficient throughout their careers. Conversely point-guards without an outside shot are typically terrible as freshman, and while they improve they rarely match the efficiency of other players:.
I define a player as a Three Pt Shooter if he takes more than 4 threes per 40 minutes played in his career. The definition is based on attempts, not makes. I define a player as a Non-Shooting PG if he earns at least 4 assists per 40 minutes played in his career. This assist cutoff was chosen to be restrictive enough to exclude players like Henry Sims, but that also means it excludes some guards who are typically viewed as point guards.Predicting the NBA 2K Players Tournament and NBA H-O-R-S-E competition champs - The Jump
I also broke the data down to look at combo guards, guards who have an outside shot and set up their teammates. It turns out that passers who have three-point range tend to have similar efficiency to spot up three point shooters and thus I grouped all three point shooters together in the table. This table is one of the key reasons people have criticized ORtg. Players that are spot-up three point shooters tend to be more efficient even though they might not be the most valuable players on the floor.
Most analysts handle this by concluding that ORtg is a stat that requires context such as Usage rates in order to interpret it. Teams as varied as Florida and Wisconsin have proven that by taking and making a bunch of threes you can have an efficient offense.Every NBA team is constantly searching for an edge, and with the success of analytics in other sports, such as Major League Baseball, NBA teams are looking to advanced technologies like machine learning and artificial intelligence AI to gain a competitive advantage.
There are many applications for AI within sports organizations, including sales and marketing, merchandising, chatbots, computer vision, and wearable tech, but this blog post focuses on a particular application — predicting player performance.
There are many ways to assess player performance. At the most basic level, basketball is about scoring more points than the opponent, so naturally points-per-game is a nice place to start. But, there are other methods to quantify player performance, and some of them get quite complex like Box Plus Minus or Player Efficiency Rating. Game Score was created by John Hollinger to give a rough measure of a player's productivity for a single game. Each modeling approach, called a blueprint, is fit on a portion of the training data and ranked by accuracy using the out of sample validation data.
In minutes, hundred of models are built, analyzed and displayed on the Model Leaderboard see Figure 1 for further evaluation. Figure 1: Top 5 models sorted by accuracy on the out of sample validation data. The Leaderboard displays badges, tags, and columns that provide information to quickly identify the model and scoring information. The lift chart is easy way to quickly evaluate the accuracy of a particular model on the dataset. The lift chart bins and sorts the expected player performance as predicted from lowest to highest.
Overlayed are the actual values for player performance. In this chart, the actual player performance values track nicely with the predicted values with a well defined slope to resolve the top performers from rest. These tools enable easy and effective communication of insights and expected performance.
One of the most important skills of a top data scientist is being able to tell a compelling story that drives decisions throughout the organization.
Here are a few of the tools available in DataRobot for model insights. Feature impact is a tool that ranks each variable in the dataset by its relative importance. This is a good tool for understanding and explaining what features, or variables, the model has determined to be most important for making accurate predictions.
Figure 3: Feature Impact. DataRobot seamlessly mines and processes unstructured text to extract information that may be predictive. DataRobot visualizes the salient information using a word cloud see Figure 4. First, the larger the word or phrase, the more often it shows up in the dataset. Second, red is associated with high performance and blue is associated with poor performance.
The more intense the colors, the stronger the association is. Because of the complexity of many machine learning techniques, models can sometimes be difficult to interpret directly. One of the strongest predictors of future performance is the minutes played recently.
The relationship between minutes played recently and future performance is quite strong.Redbull has a huge following thanks to their high-quality, visually captivating content Oreo uses video content on Facebook to engage with followers in a unique way, inviting them to play a game 5. Way too much text happening here. Petco does a great job of incorporating a seasonal element into their alr projector screen uk photo 7.
Mei Mei's Street Kitchen tags organizations they are teaming up with in posts 10. Searching hashtags can help you discover competitors, generate ideas, monitor industry conversations, and more 11. Incorporate questions and surveys into your social posts 13. Joining in on weekly movements can get your social posts some much needed publicity 14. Matador Network, utilizing quotes in their Facebook ads 19.
Utilize automated meme generators for something big and bold 20. Find technicolor tg789vac v2 flash how you're REALLY doing in AdWords.
Watch the video below on our Free AdWords Grader: Visit the AdWords Grader. Comments David Weightman I really love the tips that you have shared Megan. Megan Marrs Thanks David, I'm really glad you enjoyed the article. Megan Marrs Thanks Serge. Jackie Hi MeganThanks so much for putting into clear and easy language what I should be doing on Facebook. Jackie George All very useful stuff. Thanks for sharing Megan. George Elena Wow, what a precious collection of valuable tips.
Megan Marrs You're welcome Elena. Michael Bian nice tips. Im glad I was able to read this. Phil Ryan Great list guys.
We’re Predicting The Career Of Every NBA Player. Here’s How.
Megan Marrs Video Converter I learned a lot from you post. Rae You actually make it seem so easy with your presentation but I find tuis matter to be reallly something which I think I would never understand.
Victorino Abrugar Wow Megan. Megan Marrs Aw, thanks Victorino, I really appreciate that.