Dota 2 match analytics

Model predictions, probabilities and accuracy on real matches

Prediction accuracy

71%

63 of 89 matches

Calibration (Brier)

0.208

the lower, the more honest the probability

Accuracy on favorites

77%

37 of 48 confident

Average favorite odds

1.53

fair, without margin

How we calculate. For each match the model estimates the win probability from player ratings (player-Elo), Glicko-2 rating, form and team experience — using data before the match starts. From the probability we get the "fair odds" (1 / probability): the break-even threshold for a bet. Accuracy and calibration (Brier) are computed on real finished matches. Matches marked value are those where the market odds are above fair, i.e. the bet has positive expected value (edge = odds × probability − 1).

Betting strategies: historical backtest → arrow_forward

Your strategy: Value: edge ≥ 5% · 2% of bankrollits recommendations show up on match pages

All Live · 5 Upcoming Past Prediction hit

Upcoming and live

5

DREDreamLeague Season 30. Закрытые квалификации

odds 2.13fair

odds 1.89fair

favorite · Zenithconfidence 53% · fair odds 1.89updating live

DREDreamLeague Season 30. Закрытые квалификации

Team Resilience

odds 1.96fair

Yakult Brothers

odds 2.05fair

favorite · Team Resilienceconfidence 51% · fair odds 1.96value +5% · Yakult Brothers (odds 2.15)updating live

DREDreamLeague Season 30. Закрытые квалификации

odds 3.15fair

odds 1.46fair

favorite · Shadow Danceconfidence 68% · fair odds 1.46updating live

DREDreamLeague Season 30. Закрытые квалификации

odds 1.84fair

odds 2.19fair

favorite · CyberHeroconfidence 54% · fair odds 1.84value +46% · Kalmychata (odds 3.20)updating live

STRStreamers League #3

odds 2.52fair

odds 1.66fair

favorite · Nix Teamconfidence 60% · fair odds 1.66value +7% · Nix Team (odds 1.78)updating live

Dota 2 match analytics and predictions

This section gathers our model's predictions for Dota 2 matches: each team's chance to win the series, the fair odds and a draft breakdown. The prediction is built from pre-match statistics — player ratings (player-Elo), teams' Glicko-2 rating, current form and accumulated experience.

For finished matches we show the original probability and the actual result, and above it — honest quality metrics: prediction accuracy and calibration (Brier). "Fair odds" equal one divided by the probability and mean the threshold above which a bet has positive expected value.