• Dataset: Microsoft GeoLife Trajectory Dataset (1.2M+ points).
• Core Model: XGBoost Regressor.
• Accuracy: 0.28 RMSE (Root Mean Square Error) on a 100-point scale.

🧠 The Science: Feature Engineering

The model's high accuracy stems from integrating established sports science and physiological research:

1. Naismith’s Rule (Elevation Equivalence)

Using the 1892 Naismith’s Rule, the engine calculates Effective Distance (ED). It treats every 120 meters of vertical ascent as 1 kilometer of flat walking.

Formula: $ED = \text{Distance (km)} + \frac{\text{Elevation Gain (m)}}{120}$

2. Thermal Stress (Galloway & Maughan)

Based on environmental physiology research, the model applies a non-linear penalty for temperatures outside the "Optimal Performance Zone" (10°C – 20°C).

• Heat Strain: $+1.3%$ fatigue per degree above 20°C.
• Cold Strain: $+1.0%$ fatigue per degree below 10°C.

3. Langmuir’s Group Corrections

Adjusts the pace and fatigue accumulation based on group size, accounting for social pacing and the "slowest member" rule in group expeditions.

🛠️ Technical Stack

• Languages: Python (Data Science Pipeline).
• Libraries: XGBoost, Pandas (Vectorized processing), NumPy, Scikit-learn, Matplotlib.
• Data Sourcing: Open-Meteo Historical Weather API integration.
• Architecture: Planned integration with Next.js & Supabase via a FastAPI microservice.

📊 Model Performance

The model was optimized using Gradient Boosting with early stopping to prevent overfitting.

Metric	Value
Best RMSE	0.2828
Optimization Rounds	199 Iterations
Top Predictor	Total Distance / Elevation Gain