Model Insights

Live diagnostics for the Model90 Match Engine — SHAP feature importance, CLV, and stacker architecture.

0.1675
Avg Brier Score
lower = better (0.222 = random)
60.4%
Avg Accuracy
correct outcome picked
meta_v1
Model Version
latest deployed

Feature Importance (SHAP)

Mean absolute SHAP value from XGBoost — which signals drive predictions most.

elo diff
0.1791
elo home
0.1268
form points home 5
0.1136
elo away
0.1061
market implied home win
0.0750
form ga away 5
0.0562
poisson defense home
0.0466
days since last home
0.0429
poisson attack away
0.0408
form ga home 5
0.0348
market implied away win
0.0300
form xg for home 10
0.0288
poisson attack home
0.0279
form gf home 5
0.0257
form points away 5
0.0226

Stacker Weights

Contribution of each base model to the final ensemble output.

xgboost
40.3%
lightgbm
33.6%
random forest
26.1%

Closing Line Value (CLV)

How much the model's pre-match probabilities beat the closing bookmaker line.

No CLV data yet — fixtures need closing odds.

Model Performance & Calibration

Per-round Brier score, log-loss, RPS, ECE calibration, and detailed accuracy metrics.

View Charts →
Model Insights | Model90