Formula 1 race and championship prediction with Machine learning

Two pre-event machine-learning models for Formula 1, built only on data known before the event.A pre-race model estimates the probability that a driver finishes on the podium (CatBoost with anti-leakage features, time-decay weighting and isotonic calibration), evaluated with precision@3against the qualifying grid. A full-season model estimates the drivers' championship winner fromthe standings after each round, calibrated with softmax and temperature scaling. The work reports honest evaluation: bootstrap confidence intervals, an ablation study, and a calibration comparison against bookmaker odds. Code and trained models: Repository