We specialize in data science for sports, providing an innovative and unique mathematical system for predicting player outcomes, as well as other variables of interest.
Our approach handles complexities that are beyond the reach of conventional prediction models, but unlike machine learning it is transparent, adaptive to new prediction tasks, and theoretically justified.
ABOUT US
NEW RESEARCH
Predicting NBA Draft Pick Performance
Learn how relevance-based prediction identifies the uniquely optimal combination of previously drafted players and predictive variables for each prediction task, and how the measure of fit offers guidance on confidence and commitment to each draft prospect.
A NEW SYSTEM FOR
ASSESSING THE FUTURE
RELEVANCE-BASED PREDICTION
Based on award-winning research, the concept of statistical relevance offers a combination of transparency and adaptability that goes beyond other prediction systems.
Relevance
Relevance is a mathematically precise and theoretically grounded measure of the importance of an observation to a prediction.
Relevance is composed of two components, similarity and informativeness, both measured as Mahalanobis distances.
Fit
Fit measures the alignment of outcomes across all pairs of observations that inform a prediction task.
It gauges the confidence we should have in a particular prediction task, and it helps us determine the uniquely optimal combination of predictive variables and observations for each task.
Codependence
Conventional prediction models mistakenly assume that a set of predictive variables is equally effective across all observations, and that one set of observations is best for all sets of predictive variables.
We use fit as a joint function of variables and observations to optimally blend the best information for each predict task.
GET A COMPETITIVE EDGE
ADVANTAGE
EXTENSIBLE
Form predictions for as many outcomes as you like, considering hidden complexities that are beyond the reach of conventional prediction models.
TRANSPARENT
Identify the most relevant observations and most effective predictive variables that go into each prediction task for easy comparison with intuition, scouting reports, and more.
NUANCED
Measure the unique reliability of each specific prediction, rather than rely on a regression's R-squared which assumes all predictions are equally reliable.
ADAPTIVE
Automatically update the prediction process when circumstances change, rather than rely on machine learning which must construct a new model when faced with unprecedented circumstances.