KHL Prediction Accuracy Sign in
All project features are currently free Why the project runs on donations →
Open model evaluation

KHL prediction accuracy

One latest pre-match version per game. This page evaluates model probabilities; the selected published picks and skips have their own results ledger.

Selected picks and skips →
69.0%42 evaluated games

Model and baseline

Validated games
42
Majority-class baseline
64.3%
Lift, percentage points
4.7
Brier score
0.2219

The baseline always selects the most common outcome in this same sample. Lower Brier scores indicate better probabilities, including confidence in incorrect predictions.

Accuracy by market

Model outcomes at the 50% threshold
MarketCorrect / evaluatedAccuracy
Winner incl. OT/SO29 / 4269.0%
Over/Under 5.5 in regulation2 / 366.7%
Period 1 · Over/Under 1.528 / 4266.7%
Period 2 · Over/Under 1.520 / 4247.6%
Period 3 · Over/Under 1.519 / 4245.2%

Accuracy by confidence

ConfidenceGamesPredicted probabilityActual accuracy
70%+0
60–69%1564.0%80.0%
55–59%1458.4%50.0%
50–54%1352.7%76.9%

Winner accuracy by team

TeamCorrect / evaluatedAccuracy
Локомотив6 / 785.7%
Металлург Мг6 / 785.7%
Авангард4 / 757.1%
Ак Барс3 / 742.9%
Салават Юлаев5 / 683.3%
Торпедо5 / 683.3%
ЦСКА4 / 666.7%
Динамо Мн0 / 50.0%
Амур3 / 3100.0%
Лада3 / 3100.0%
СКА3 / 3100.0%
Нефтехимик2 / 366.7%
Драконы1 / 333.3%
Адмирал2 / 2100.0%
Динамо М2 / 2100.0%
Северсталь2 / 2100.0%

Prediction accuracy is measured in games involving each team; the model can also choose its opponent. Small samples are unstable.