DARKO, aka Daily Plus-Minus or DPM, is created by Kostya Medvedovsky. DARKO (stands for “Daily Adjusted and Regressed Kalman Optimized“) is a composite predictive metric that uses box score and plus-minus stats. By blending these components in proportion to the number of total possessions of a player, recent performance is weighted more heavily to better predict future game outcomes.
Their DARKO.app updates daily for every player by collecting publicly available data to account for time series and sample size using the complex methods of “exponential decay” and “Kalman filters”. DARKO.app website (web presence developed by Andrew Patton) also offers great data visualization tools.
The main difference between DPM and the other player impact metrics is that DPM solely looks “forward” by giving the results in a Bayesian model which projects all elements of the box score. Backwards-looking in GOAT debates are not meaningful with DPM.
Based on root mean square error, DPM is better than any all-in-one metric in terms of predictive analytics where EPM ranks second and LEBRON third.
The results are below, and are sorted by RMSE. I've also shown mean average error and R^2. Different stats lead in the various metrics. EPM especially stands out to me here for leading in MAE and R^2. 7/n pic.twitter.com/6QeWBm32sW
— Kostya Medvedovsky (@kmedved) January 30, 2021