Introduction - Scenario Engine and Reporting Platform for Racing Queensland
PAG produced a scenario engine and reporting platform to predict turnover for racing events across Queensland.
Scheduling an event to maximise public participation and revenue is a complex task. This complexity dramatically increases when organisers need to schedule events across Queensland for a complete year in advance.
While years of industry experience provide schedulers with a level of intuition, the task of scheduling an average 32 race meetings per week, over 3 codes (thoroughbred, harness and greyhounds) is substantial. While understanding the variables that lead to a successful met may be domain knowledge, the elements that combine to provide the optimal calendar are not readily programmed by individuals.
In addition, schedules need to be fair, providing the 120 race clubs, located across the state, and their communities with an opportunity to make their individual events successful. Schedules that make single events highly profitable at the expense of smaller clubs are not acceptable or desirable for the industry or the economy of regional, rural and remote areas.
Project Info
Client
Racing Queensland
Service
Scenario Engine and Reporting Platform
Industry
Sports/Racing
Challenge
Racing Queensland (RQ) have a racing schedule that is timetabled in advance, however, with flexibility to change parameters within events, such as start time, prize money and Sky Racing broadcasts. In addition, other factors such as track condition are often known in the short term, rather than at the time of scheduling. A scenario tool that could accurately inform schedulers of the likely impact any changes in these variables would have on event turnover was required for them to make data informed decisions.
A data-driven approach also allowed schedulers and code coordinators to work together for the benefit of the industry, the individual codes and the racing clubs in securing a vibrant future for this sport.
Solution
PAG collaborated with schedulers and code managers from RQ to obtain a list of suggested variables that could potentially effect the turnover at individual race meetings. The initial list of candidates was around 75 variables.
Historical data was used to develop Machine Learning algorithms to forecast turnover at an event level. The methods developed accurately predicted turnover, for each event on the racing calendar over an 18 month out-of-sample window.
PAG provided RQ with a series of visualisations within a companion dashboard to allow the end-users to interpret the importance of the variables to the outcome at an overall, code and race level. The dashboard also provided tabs for the forecast turnovers, which updated through the window to highlight disparities, which could then be reconciled with conditions on the day of the event and other external factors.
Outcome
The scenario engine is a web-based interface, and was embedded into Microsoft Power BI, which was the client’s preferred dashboard platform. The engine predicted turnover at the meeting level and was used by management to provide evidence-based decision making. The application enabled users to trial different scenarios with the goal of optimising turnover by understanding scheduling criteria.
The backend of the application was deployed on a cloud platform within a secure, isolated network environment. Infrastructure provisioning was automated using infrastructure-as-code tools, ensuring consistency and scalability. The system was designed to support automated retraining of machine learning models within a serverless compute environment. Trained models were stored in cloud-based object storage and dynamically retrieved by the serving component, which operated within a lightweight serverless framework. A managed API layer facilitated communication between the serving engine and the web-based frontend interface used by staff.
As the machine learning models were re-trained automatically, a model monitoring dashboard was setup to ensure there was no unexpected degradation in model performance. PAG provided training and documentation for both the business users of the product, and to the IT staff who maintain the system. over.
Impact
The client has a scenario engine that is equipped with an automatic retraining model facility by which to predict event outcome and make data-driven scheduling decisions. The model is set to re-train, as data collection continues to expand, and conditions are likely to evolve. This approach extends the life of the product for our client.
RQ management and schedulers are able to test the viability of making schedule changes and determine if they are advantageous or harmful to the overall racing program. The scenario and companion dashboard allow complex interactions to be modelled and presented in a user friendly environment.
