A Data-Centric Approach to Weather Prediction: IoT and Machine Learning on Drone Platforms

IoT and Machine Learning on Drone Platforms

📄 Abstract

This research presents a drone-based weather prediction system that leverages IoT sensor networks and machine learning algorithms to optimize agricultural decision-making. The approach combines real-time sensing, edge computing, and pre-trained ML/DL models to achieve high-accuracy weather forecasts in remote agricultural areas, minimizing latency and enhancing responsiveness.

---

🔍 Problem Statement

Modern agriculture demands precise environmental monitoring. However, low-cost IoT devices often lack the computational capabilities required for real-time predictions using traditional machine learning techniques. Cloud computing introduces latency, making timely actuation difficult in critical scenarios like rainfall or frost warnings.

---

🚁 Proposed Solution

We propose a hybrid edge-cloud framework where:

• Historical weather data is used to train ML/DL models (Logistic Regression, XGBoost, CatBoost, etc.).
• These models are deployed on edge nodes mounted on drones to reduce latency.
• IoT sensors capture real-time environmental data which is processed onboard by the drone.
• The system provides real-time actuation decisions for agricultural automation (e.g., irrigation, pesticide spraying).

---

🧠 Machine Learning Models Evaluated

Model	Performance (Australia Dataset)	Performance (Bikaner Dataset)
Logistic Regression	79.5% Accuracy	91.3% Accuracy
Decision Tree	86.1% Accuracy	88.3% Accuracy
Random Forest	92.7% Accuracy	90.8% Accuracy
XGBoost	94.96% Accuracy	90.3% Accuracy
CatBoost	93.9% Accuracy	91.2% Accuracy
LightGBM	86.6% Accuracy	90.8% Accuracy

📈 Evaluation Metrics: Accuracy, Precision, Recall, F1-score, Cohen’s Kappa, AUC-ROC.

---

🌍 Datasets Used

• Rain in Australia (Kaggle)
• Bikaner (India) Weather Data: Collected from NASA POWER Project and IMDLib

---

⚙️ System Architecture

1. Cloud Layer: Train models on historical weather data.
2. Edge Layer (Drone): Deploy pre-trained model for real-time inference.
3. IoT Layer: Sensors collect temperature, humidity, pressure, etc.
4. Actuator Layer: Executes automated decisions on the field.

---

🔬 Experimental Results

• XGBoost achieved the highest prediction accuracy (94.96%) on the Australian dataset.
• Logistic Regression performed best on the Bikaner dataset.
• Drone-mounted edge nodes demonstrated real-time capability with low-latency decision-making.

---

🧾 Citation

If you find this research helpful, please consider citing:

@INPROCEEDINGS{10984355,
  author={Parihar, Devendra and Chaudhary, Ajay and Peddoju, Shreehitha and Kadarla, Kavitha},
  booktitle={2024 IEEE India Geoscience and Remote Sensing Symposium (InGARSS)}, 
  title={A Data-Centric Approach to Weather Prediction: IoT and Machine Learning on Drone Platforms}, 
  year={2024},
  volume={},
  number={},
  pages={1-4},
  keywords={Machine learning algorithms;Accuracy;Computational modeling;Decision making;Weather forecasting;Machine learning;Prediction algorithms;Sensors;Internet of Things;Drones;Internet of Things;IoT;Edge Node;Drone;Weahter Predition;Weather Forecasting;Precision Farming;Smart Agriculture;Machine Learning},
  doi={10.1109/InGARSS61818.2024.10984355}}

---

🤝 Contributions & Acknowledgements

• Devendra Parihar – AI Developer at Helios Solutions, Vadodara, India
• Ajay Chaudhary – Engineering College Bikaner, India
• Shreehitha Peddoju – NIT Rourkela
• Kavitha Kadarla – COER University, Roorkee

Special thanks to datasets from NASA, IMD, and Kaggle.

---

📬 Contact

For queries, reach out to: 📧 devendraparihar340@gmail.com