F1 Kinetic

Probabilistic race forecasting and strategy optimisation engine built on 70+ years of Formula 1 telemetry (1950–2023).

---

Overview

Component	Description
Probabilistic Forecasting Engine	Bayesian Inference + Feature Engineering to predict race outcomes; achieves a 15% improvement in outcome reliability over the grid-position baseline
Monte Carlo Simulation Framework	10,000+ iterations per race across 23+ Grand Prix to optimise fuel-load and pit-stop strategy through multi-dimensional data modelling
Exploratory Data Analysis	Historical win counts, speed trends, and grid-position correlation across the full 70-year dataset

---

Notebooks

Notebook	Purpose
probabilistic_forecasting.ipynb	Bayesian Beta-Binomial prior, 11-feature engineering pipeline, Random Forest ensemble, reliability benchmarking
monte_carlo_simulation.ipynb	Lap-by-lap race simulator, fuel-load sensitivity analysis, pit-stop window optimisation for 23 GPs
F1_proj.ipynb	Exploratory data analysis — driver/constructor win counts, speed facet grids, grid-vs-finish regression

---

Probabilistic Forecasting Engine

Predicts podium finishers using a Beta-Binomial Bayesian model updated with grid-position likelihood ratios, then ensembled with a Random Forest trained on 11 engineered features:

career_win_rate  |  career_podium_rate  |  recent_form_5
circuit_avg_position  |  constructor_win_rate  |  constructor_recent_points
grid_position  |  dnf_rate  |  experience  |  podium_rate  |  quali_time_delta

Result: ~15% improvement in per-race podium hit-rate over the grid-position-only baseline (evaluated on 2018–2019 hold-out seasons).

---

Monte Carlo Simulation Framework

Each iteration samples from physically-grounded distributions:

Variable	Distribution
Fuel load	Uniform(85, 110) kg
Pit stop duration	Normal(25 s, 2 s)
Lap time noise	Normal(μ_circuit, σ_circuit)
Fuel burn penalty	0.03 s/lap per kg remaining
Safety car events	Poisson(λ = 0.25) per race
DNF probability	Per-circuit historical rate

10,000 iterations × 23 Grand Prix = 230,000+ simulated race scenarios

Outputs per GP: optimal fuel load (± confidence), best pit-stop strategy, pit-window lap ranges, and 1-stop vs 2-stop time delta.

---

Tech Stack

• Python — core language
• Pandas — data ingestion, merging, and rolling feature engineering
• Scikit-Learn — StandardScaler, LogisticRegression, RandomForestClassifier
• SciPy — Beta distribution (Bayesian inference), statistical utilities
• NumPy — Monte Carlo sampling, vectorised simulation
• Matplotlib / Seaborn — visualisations

---

Dataset

Formula 1 World Championship data, 1950–2023 (Ergast format).

File	Contents
results.csv	Race results, finishing positions, fastest laps
races.csv	Race calendar, circuits, dates
drivers.csv	Driver registry
constructors.csv	Constructor/team registry
qualifying.csv	Q1/Q2/Q3 session times
circuits.csv	Circuit metadata

Dataset source — Kaggle

---

Screenshots