This project aims to develop a machine learning model to accurately predict caloric expenditure during physical activities using physiological metrics such as heart rate, body temperature, activity duration, weight, and age. By leveraging advanced regression models, we enhance accuracy over traditional estimation techniques.
- Multi-Model Approach: Implements Linear Regression, Decision Tree, Random Forest, and XGBoost for comparative analysis.
- High Accuracy: XGBoost achieves a 35% reduction in Mean Absolute Error (MAE) compared to baseline models.
- Optimized Performance: Feature selection and hyperparameter tuning lead to a 40% improvement in computational efficiency.
- Scalable & Adaptable: Easily integrates with fitness applications and wearable devices.
- Size: 15,000 records
- Link: https://www.kaggle.com/datasets/fmendes/fmendesdat263xdemos/data
- Features: Heart rate, body temperature, duration, age, weight, height, and activity type
- Preprocessing:
- Handling missing data with imputation
- Normalization for feature scaling
- Feature selection to enhance efficiency
- Programming Language: Python
- Libraries: NumPy, Pandas, scikit-learn, XGBoost, Matplotlib, Seaborn
| Model | MAE Reduction | R² Score |
|---|---|---|
| Linear Regression | 12% | 0.75 |
| Decision Tree | 18% | 0.81 |
| Random Forest | 28% | 0.88 |
| XGBoost | 35% | 0.92 |
- Data Collection & Preprocessing: Cleaning, feature engineering, normalization.
- Model Selection & Training: Implementing multiple regression models.
- Hyperparameter Tuning: Using GridSearchCV to optimize performance.
- Evaluation & Comparison: Measuring performance with MAE and R².
- Deployment (Future Scope): Integration with real-world fitness applications.
- Implement deep learning models for further accuracy improvement.
- Deploy as an API for real-time calorie tracking in fitness apps.
- Extend dataset to include diverse demographics for improved generalization.
Want to contribute? Feel free to open an issue or submit a pull request.