Part 1: Can Analytics help Talent ID and Quantify Creativity in Sport?

6 Feb

This is the first in a series of blogs I’ll be writing for the Sports Analytics Innovation Summit on London which is one of the leading meet ups in the world.

Analytics is Child’s Play

Analytics at Work

Anyone who has played or coached sport at any level will have constantly analyzed what theirs and their opponent’s next move is.  A perfect example of this is watching a group of 6 year olds play cat and mouse.  At first both the cat and mouse run round in circles to catch each other, then suddenly the cat will make a prediction based on the information gained form the game: “if I change direction and go the other way will that increase my chance of catching the mouse?” (or words to that effect) and sure enough it works.  Suddenly the 6 year old has started to analyze the concepts of the game based on the knowledge they have (data) and then develop a new strategy to affect a future outcome.  In a sense they have created a feedback loop which will become more advanced with the more data they consume.  This is probably the simplest form of sports analytics and should be the basis of the work we carry out when working with coaches and players.

Keep it Simple and Make it Meaningful

Whether as a fanalysist, researcher or team principle, the amount of data available to analyze performance is eye watering.  For a 90 minute game of football Prozone capture 54,000 data points by tracking each player’s individual movement through a multi camera system.  If you think that’s impressive then IBM’s Slam Tracker uses 39 million data points from 7 years’ worth of Grand Slam tournaments to determine player patterns to predict the outcome of the match.

However it’s not the level of data you collect that’s important but how it’s applied to gain an advantage.  Two excellent recent examples of how to use data to gain an advantage are British Rowing and Leicester Tigers.

British Rowing are without doubt one of this country’s biggest sporting success stories.  In 2008 they won their highest amount of medals at an Olympic games and where the most successful nation. However they knew that to keep ahead of the game they need to progress further.  Enter sports analytics.  To better understand the key physical, mental, tactical and technical factors that underpinned racing success British Rowing undertook a detailed factor analysis.  Based on data collected over the decades from past Olympic cycles they were able to first identify the factors which correlated to winning and through regression analysis developed a training program targeted.  The rest is now history with the team breaking their record medal haul winning 4 gold, 2 silver and 3 bronze at London 2012.

Leicester Tigers recognized that to be able to compete in both the league and European competition it was vital for them to be able to get their best players starting as many games as possible.  They identified the largest preventative factor to this was the occurrence of injuries.  Along with increases in size and physicality of players the number of injuries suffered be players also increased, therefore they had to think smartly about how to reduce the number of injuries being suffered in a season.  A partnership with IBM was struck with the idea of preventing injuries by predicting when they were likely to happen.  By using IBM predictive analytics the clubs sports science team where able to collate all the data collected on players from training load, sleep patterns to past injuries.  Based on past data they are able to input current information through the predictive software to model the risk percentage a certain player is to become injured.  The club are currently 3rd in Guinness Premier League and top of their Heineken Cup group respectively.

TID and Measuring Creativity

So what about the future?  One of the largest area’s which analytics can support is the golden goose of talent identification (TID).  It could be said that a number of sports have extremely inefficient talent identification systems in place.  Football clubs invest up to £5m per year into an academy which they may recoup through the sail of youth products or through savings on not having to pay for new player.  So in terms of a balance sheet academies can and generally do stack up, however if we look at this from a talent conversion quota perspective the story is not so rosy.  The number of academy players being released before the age of 16 is 50% and for those fortunate to progress through to u18 level on average 93 per season gain a 1 year contract.  This works out an average of 4.65 players per team and gives us a 25% success rate for the players who are left in the system.

Young Iniesta

So how can analytics help to solve this?  UK Sport and the English Institute of Sport have already set the ball rolling with several highly successful TID programmes which have focused on converting athletes who may not have progressed at their primary sport but have the potential to transfer and excel at another.  By developing a series of physical performance indicators the programmes have been able to identify a number of athletes who have gone on to win medals at world level.

Sport isn’t just about physicality, it also requires a high level of sporting intelligence.  With most team sports looking to unearth the next Lionel Messi, Tom Brady or Kobi Jones.  So how do we ensure we identify and nurture the right talent from raw potential to genius?

Ground breaking research by Andreas Grunz, Daniel Memmert and Jurgen Perl has looked at quantitatively capturing game intelligence.  By applying artificial neural networks (ANN) to learn repeated behaviours, Memmert and Perl have develop a Dynamical Controlled Network (DyCoN) which can be used to capture unique movements or solutions to pre-set problems.  Where normal ANN’s are trained to only recognize frequent actions (i.e. the norm) ignoring deviations from the norm (i.e. creativity), a DyCoN ANN can be trained to recognize infrequent but important changes from the norm.

So How Does This Work?

Put very basically a sequence can be converted from film to a series of X and Y coordinates with specific coding to describe actions such running, throwing catching etc… This data is then fed into the DyCoN which is trained to recognize patterns of interactions as described above.  This then allows actions to measured and mapped out creating a neural network of all the actions observed. The system can the identify unique responses which indicate creativity.

Example of a trajectory containing a creative neuron (grey circle), representing a rare and adequate action. From ANDREAS GRUNZ, DANIEL MEMMERT, JÜRGEN PERL

Example of a trajectory containing a creative neuron (grey circle), representing a rare and adequate action. From ANDREAS GRUNZ, DANIEL MEMMERT, JÜRGEN PERL

From this it is possible to identify what may be considered the normal response and what a unique solution is.  While the research is still in its infancy it does shed light into a new area that sports analytics can be used for.

As previously mentioned this is the first in a series of blog for the Sports Analytics Innovation Summit. I will be building on this concept of quantifying creativity and introducing you to more of the research in hopefully an understandable way.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: