LightGBM model trained on 690 matches across 104 venues. 63 features. 93.5% validation accuracy on past World Cups.
A LightGBM classifier trained on 690 international T20 matches, using a layered feature group approach to prevent overfitting.
690 international women's T20 matches across 104 venues. Match outcomes, toss results, and venue statistics all captured.
63 features across 4 kept groups: team strength, venue fit, phase-by-phase batting/bowling, and matchup/toss. A 5th group (G4: fielding, captaincy, pressure) was tested and dropped — it increased log-loss and was excluded.
Feature groups added one at a time. A group is only kept if it reduces log-loss on the validation set — no data leakage.
Held out the entire 2022 WC (South Africa) and 2024 WC (Bangladesh) as test sets — never seen during training or tuning.
Gradient boosted trees with early stopping, L1+L2 regularisation, and shallow depth (15 leaves) to prevent overfitting on 644 rows.
For each 2026 match, features are computed using real venue data. The model outputs win probability for both teams.
Feature Importance
Top 20 features by LightGBM split importance. Head-to-head record and venue chase win rate are the strongest signals.
The model was tested on two complete past World Cups it had never seen during training. Each dot below represents one match.
Missed: Sri Lanka beat South Africa (opener upset); South Africa beat England (semi)
Only miss: South Africa beat Australia in group stage (52% Australia predicted)
Australia and England predicted to top their groups and advance to the semi-finals.
Edgbaston · Old Trafford · Headingley · Rose Bowl · Bristol · Lord's
| # | Team | Pts |
|---|
Edgbaston · Rose Bowl · Headingley · Bristol · Old Trafford · The Oval · Lord's
| # | Team | Pts |
|---|
30 group stage matches shown here. 3 knockout matches (SF1, SF2, Final) are shown in the Bracket section — 33 total predicted.
Win probabilities for every group stage match. Filter by group or confidence level.
Confidence bands: HIGH = win prob ≥75% · MED = 60–75% · LOW = <60% (close call)
Australia predicted champions at Lord's on July 5.
Semi-Final 1 · Jun 30 · The Oval
Semi-Final 2 · Jul 2 · The Oval
Final · Jul 5 · Lord's
ELO ratings computed from all historical results, weighted by recency and opposition strength. Australia lead by a significant margin. Note: Ireland rank 3rd in ELO (1240) due to strong bilateral series results, but are predicted to finish bottom of Group B — ELO alone doesn't account for venue familiarity or bowling matchup deficits against top-6 opposition at English conditions.
Team Profiles
Select a team to see their phase-by-phase strengths. Scores normalised 0–1.
All-time head-to-head win rates in international T20 cricket. Green = strong record, red = weak. Hover a cell to see details.