prediction models

There’s no such thing as a perfect prediction model (Sketchnotes)

The modern world is increasingly enamoured by the possibilities of big data and sophisticated analytics. In sport, the application of data analytics continues to rise rapidly as measurement technologies and analysis platforms become more advanced yet accessible. Indeed, a prevailing view is that if you’re not using data analytics to inform decision making at every level of a sporting organisation, then you are falling behind:

 

So how close are we to the holy grail of being able to accurately predict performance, illness, injury?

I’d argue that we are a long way off. Here’s why: attempting to forecast future events is not a new endeavour. Take meteorology, for example. We have accurate and reliable measures of weather patterns collected daily for years and even decades, yet the weather forecasts we see on the news every night rarely extend beyond 7 days. Predictions of minimum and maximum temperatures have a high degree of accuracy – the MetOffice (United Kingdom) achieves ~85% and 90% accuracy for these predictions, respectively – but there remains a degree of error. Predictions of rain are less accurate, and predictions of uncommon events such as earthquakes are considerably less accurate again.

In sport, we face a whole host of challenges, chief among them being the quality of our measures, the depth (or lack thereof) of historical data using consistent measures, and the difficulty of developing models to explain highly variable events that may not occur frequently.

Jacquie Tran - "There's No Such Thing As A Perfect Model" (Sketchnotes)

Click to view full size

But as always, challenges present opportunities. Improvements in any one of these areas brings us closer to that proverbial holy grail.

Do you work in sports analytics as a researcher or applied scientist? What are your thoughts about our capacity to predict athlete outcomes now and in the near future?

Advertisements