Prediction Model

Overview

Two separate models run per race weekend: one for the grand prix, one for the sprint (sprint weekends only). Both use weighted feature scoring + softmax. The features and weights differ because sprint races are a fundamentally different format (~17 laps, no pit stops, track position dominates).

Prediction standings

Race prediction detail

Where the data comes from

Every feature score ultimately originates from one of three places:

1. FastF1 (live F1 session data, Python library)

FastF1 is the sole external data source. It provides telemetry and session data from the official F1 timing feed. The Python ingest jobs pull this data after each session and store it in raw tables.

FastF1 HeadshotUrl is used for driver photos (available 2019+, not available earlier). FastF1 does not provide team logos — those are stored as static files in web/public/teams/.

2. Computed aggregates (run after each race)

compute_season_stats reads from the raw tables above and computes season-wide summaries. These are the inputs for most driver/team-level features.

3. Static / hardcoded

Some inputs never change and are seeded once into the database or hardcoded in Python.

Data pipeline — end-to-end flow

FastF1 session
    │
    ▼  (ingest jobs)
raw tables: race_results, qualifying_results, lap_times,
            sprint_results, sprint_lap_times, fp2_long_run_times
    │
    ▼  (compute_season_stats — after each race)
driver_season_stats, team_season_stats
    │
    ├── + qualifying_results (sector times, grid position)
    ├── + lap_times (tyre degradation, long run pace)
    ├── + fp2_long_run_times (FP2 stint data)
    ├── + circuits.overtake_rate  ← static
    └── + circuits.sc_probability ← static
         │
         ▼  (compute_features / compute_sprint_features)
    driver_prediction_features / driver_sprint_features
         │
         ▼  (compute_predictions / compute_sprint_predictions — softmax)
    race_predictions / sprint_predictions
         │
         ▼  (Hono API reads)
    /prediction page — displays win probabilities + feature breakdown

Race Status Flow

conventional weekend:
  scheduled → qualifying_done → completed
                   │                │
          compute_features    ingest_race
          compute_predictions compute_season_stats

sprint weekend:
  scheduled → sprint_qualifying_done → sprint_done → qualifying_done → completed
                       │                   │               │
              compute_sprint_features  ingest_sprint  compute_features
              compute_sprint_predictions               compute_predictions

Main race predictions are computed at qualifying_done. Sprint predictions are computed at sprint_qualifying_done (after SQ session).

Grand Prix Model — weighted-v3 — 12 Features

All feature scores are normalized to [0.0, 1.0] using min-max normalization across the driver field for that race.

Weighted Score

raw = (
    car_perf                  * 0.20 +
    long_run                  * 0.15 +
    tyre_deg                  * 0.08 +
    reliability               * 0.08 +
    quali_delta               * 0.08 +
    driver_rating             * 0.08 +
    win_rate                  * 0.08 +
    luck                      * 0.07 +
    sector_strength           * 0.06 +
    circuit_adj_start_pos     * 0.07 +
    circuit_adj_position_gain * 0.03 +
    weather_score             * 0.02
)

track_overtake_score is no longer a standalone additive feature — its value is baked into the circuit-adjusted starting position and position gain multipliers.

Sprint Model — sprint-v2 — 8 Features

Intentionally different from the main race model. Sprint races are ~17 laps with no pit stops — track position and car dominance matter far more than long-run strategy or reliability.

Weighted Score

raw = (
    car_perf              * 0.25 +
    circuit_adj_start_pos * 0.25 +
    short_run             * 0.10 +
    driver_rating         * 0.10 +
    weather_score         * 0.08 +
    win_rate              * 0.08 +
    luck                  * 0.08 +
    sq_delta              * 0.06
)

Sprint driver rating and win rate use sprint-specific history when ≥3 sprint races are recorded, otherwise fall back to main race stats (scaled to match the 8-point sprint maximum vs 25-point race maximum).

Softmax → Win Probability

exp_s = np.exp(scores / T)
win_probability = exp_s / exp_s.sum()

Temperature T = 0.3. Lower temperature makes the model more decisive — small score differences produce larger probability gaps. Do not increase T.

Probabilities sum to 1.0 across all drivers in a race. Same temperature applies to both the sprint and main race models.

Predicted Position

Drivers are ranked by win_probability descending. Rank 1 = predicted winner, stored in race_predictions.predicted_winner_id (or sprint_predictions.predicted_winner_id for sprints). All driver positions are stored in driver_prediction_features.predicted_position (or driver_sprint_features.predicted_position).

Data Availability by Era

For races without qualifying data, compute_features cannot run — those races have no predictions.