This project addresses sensor drift in metal oxide semiconductor (MOS) gas sensors using a Random Forest Regressor. Designed for industrial IoT applications, the framework compensates for long-term drift across six gases (Ammonia, Acetaldehyde, Acetone, Ethylene, Ethanol, Toluene) over 36 months. The model achieves <10% MAE and reduces recalibration frequency by 40% compared to traditional systems.
- Drift-Resistant Architecture: Prioritizes transient sensor dynamics (EMA features) over steady-state responses.
- Multi-Gas Validation: Tested on 13,910 measurements from 16 sensors.
- Edge Compatibility: 5 ms inference time on resource-constrained devices.
- Batch-Wise Analysis: Performance tracking across 10 temporal batches.
| Metric | Validation Set | Test Set |
|---|---|---|
| R² Score | 0.997 | 0.969 |
| MSE | 78.12 | 901.07 |
Top drift-resistant features: Sensor 106 (32.4%) and Sensor 98 (21.5%).
R² degradation from 0.98 (Batch 1) to 0.92 (Batch 10) over 36 months.
- Clone the repository:
git clone [https://github.com/yourusername/gas-sensor-drift-compensation.git](https://github.com/Saurabh-html/Saurabh-html-Gas-Sensor-Drift-Compensation-and-Concentration-Quantification/edit/main) cd gas-sensor-drift-compensation - Installation Dependencies pip install -r requirements.txt
📂 Repository Structure . ├── data/ # Dataset (raw/processed) ├── models/ # Pretrained models ├── src/ # Source code │ ├── preprocess.py # Data preprocessing │ ├── train.py # Model training │ └── evaluate.py # Performance evaluation ├── notebooks/ # Jupyter notebooks for EDA ├── media/ # Visualizations/results └── requirements.txt # Dependencies
Model Deployment
Output
🔮 Future Scope -Deploy pruned models on Raspberry Pi for real-time drift correction. -Integrate LSTM layers to capture temporal drift patterns. -Extend to multi-sensor fusion networks.
🤝 Contributing Contributions are welcome! Open an issue or submit a PR for:
-Bug fixes -Performance optimizations -New features (e.g., hybrid models)
📜 License MIT License. See LICENSE for details.