HOW REAL-TIME, PLAYER PERFORMANCE DATA IS DRIVING MATCH PREDICTION
Data-driven decisions are not just for work. Just ask a sports fan or look at the online sports betting industry, where low-latency, official data is driving growth.
The provision of high-profile, official data has been a key driver behind the growth of online sports betting.
Over the last decade, tennis betting in the sport market has more than doubled in Europe and accounts for the most wagers after football and horse racing
When IMG Arena wanted to better leverage large datasets using technology to deliver valuable machine learning driven insights to the business and its customers, they turned to data pioneers at T-DAB.AI (T-DAB) for guidance.
Launched in 2012, IMG Arena specialises in bringing the sports betting industry closer via its federation services and world class sporting content.
IMG Arena’s global, round-the-clock coverage delivers data from 45,000+ sporting events per year, and IMG wanted to better leverage their large sports datasets using technology to deliver valuable predictive insights for fans and bookmarkers and deliver an enhanced and personalised experience for customers.
Working in close partnership with the team at IMG, T-DAB set out to deliver predictive, real-time player and match insights to IMG users, while enabling data storytelling through a robust, scalable cloud architecture.
The project combined focused solutions design and practical machine learning driven analytical activities to explore the latest developments in sports event prediction, simulation and forecasting. The solution has been developed around delivering four prioritised goals for the user:
1- Player segmentation
This solution used clustering, an unsupervised machine learning method, to segment players by their playing style. Utilising a selection of features, such as serving efficiency and performance on court surfaces, makes it possible for users to separate grass specialists like Roger Federer from clay specialists such as Rafael Nadal.
2 – Predict winners prior to matches
Following the testing and comparison of several different supervised machine learning algorithms (classifiers), historical data was leveraged to predict winners to matches… prior to the event. And with a high level of real and potential accuracy (61-76%).
3 – Predicting winners – in-match
To predict match outcome as the match is ongoing, a similar approach to the pre-match prediction was followed that provided a prediction at every moment of the match.
While the model reached 70% accuracy, this increased after the first set to 77%. It’s an encouraging result from a minimal feature set.
4 – Predicting player momentum
To keep users immersed in the experience, the solution provides insight on player momentum throughout the match.
Momentum, in this instance is defined as the relative, instantaneous probability of winning the match. This is possible thanks to trend analysis performed on the point-by-point probability curve of a player to win the match with great success.
The machine learning development, training and testing for each phase was conducted in Python on AWS with experiments recorded using the ray[tune] library. The machine learning algorithms were mainly sourced from the scikit-learn library.
A collaborative approach
IMG acted as domain experts throughout the project with T-DAB providing technical guidance and developmental steering. T-DAB data scientists worked closely with the architecture and engineering teams and devised an actionable roadmap to ensure the best outcomes could be delivered for IMG and their users.
Throughout the development, T-DAB also conducted an architecture review to compare its current AWS infrastructure with other vendors, providing feedback on the suitability of the target architecture required to support the use cases.
At T-DAB, we recognise the role collaboration plays in a project’s success. One core requirement for the client was to build up an in-house data science team. Once established, lasting legacy services were delivered, including detailed methodology reports, as well as support from our data scientists onsite in a one-to-one setting.
Undertaking a project of this scale delivers many benefits beyond the solution itself. The project enabled IMG to:
- identify improvements in their existing cloud architecture
- deliver an analytical roadmap
- develop an internal data science capability, and
- deliver a scalable approach to drive fan engagement across multiple sports.
It also enabled them, as well as their customers, to seek new opportunities, revenue streams and markets in an industry already valued at more than $85 billion.
About The Data Analysis Bureau
T-DAB.AI is data science and data engineering innovation company. We develop innovative, bespoke machine learning-driven solutions to allow anyone to infuse technology with the spark of predictive intelligence.