Menu
⚾ Teams 📈 Markets 🏆 Playoffs 📊 Backtest 🔬 Features ℹ️ About
2026 Season

Global Feature Importance

Which features drive the ensemble model's predictions?

Top 20 Features

All Features

Rank Feature Importance Relative
1 proj_over_under 0.2287
2 proj_wins 0.1414
3 proj_win_pct 0.1056
4 pythag_wins_prev_year 0.0639
5 schedule_strength 0.0384
6 road_win_pct_prev_year 0.0344
7 whip_prev_year 0.0281
8 runs_allowed_prev_year 0.0279
9 park_adj_runs_allowed_prev_year 0.0266
10 park_adj_home_runs_prev_year 0.0235
11 home_win_pct_prev_year 0.0201
12 over_under_prev_year 0.0186
13 total_war_prev_year 0.0183
14 bullpen_war_prev_year 0.0173
15 rotation_war_prev_year 0.0157
16 wins_prev_year 0.0142
17 age_diff_prev_year 0.0140
18 home_road_diff_prev_year 0.0139
19 win_pct_prev_year 0.0137
20 wins_3yr_avg 0.0133
21 pythag_luck_3yr_avg 0.0132
22 era_prev_year 0.0131
23 home_runs_prev_year 0.0123
24 avg_age_batting_prev_year 0.0122
25 runs_scored_prev_year 0.0118
26 batting_avg_prev_year 0.0117
27 ops_prev_year 0.0116
28 bat_war_prev_year 0.0098
29 avg_age_pitching_prev_year 0.0094
30 pit_war_prev_year 0.0089
31 park_adj_runs_scored_prev_year 0.0083
32 ops_proj 0.0000
33 era_proj 0.0000
34 runs_scored_proj 0.0000
35 whip_proj 0.0000
36 pythag_wins_proj 0.0000
37 win_pct_proj 0.0000
38 over_under_proj 0.0000
39 manager_win_pct 0.0000
40 manager_tenure_years 0.0000
41 injured_war 0.0000
42 injured_pitcher_war 0.0000
43 injured_batter_war 0.0000
44 war_added 0.0000
45 war_lost 0.0000
46 net_war_change 0.0000

Feature Categories

Prior Performance

  • • Previous season wins
  • • 2-year rolling average
  • • 3-year rolling average
  • • Win % trends

Projections

  • • ZiPS projected wins
  • • Steamer projected wins
  • • Depth Charts consensus
  • • FanGraphs WAR projections

Contextual

  • • Park factors
  • • Division strength
  • • Year-over-year change
  • • Playoff status

Ensemble Composition

The predictor uses a weighted ensemble of three complementary models:

  • XGBoost (50%): Captures non-linear interactions between features (e.g., park factors × lineup strength)
  • Random Forest (30%): Robust to outliers and provides uncertainty estimates via tree variance
  • Ridge Regression (20%): Linear baseline for interpretability and stability

Feature importance shown above is averaged across all three models, weighted by their ensemble contribution.