EcoPredict is a framework designed to predict CO2 emissions in vehicular networks using Centralized, Federated, and Split Learning techniques. It processes data from Roadside Units (RSUs) to create models that help reduce emissions in urban environments.
This project generated a research paper titled "EcoPredict: Assessing Distributed Machine Learning Methods for Predicting Urban Emissions," which has been accepted for publication at the IEEE VTC 2024.
EcoPredict focuses on three different machine learning architectures:
- Centralized Learning: All data is aggregated and processed on a central server.
- Federated Learning: Data remains on local RSUs, with only model updates sent to the server, enhancing privacy.
- Split Learning: The model is split between the RSUs and the central server, balancing computation and communication.
- Multi-Model Comparison: Compare Centralized, Federated, and Split learning.
- CO2 Emission Prediction: Utilizes CNN-LSTM models to predict CO2 emissions.
- Real-Time Analysis: Tested in urban traffic scenarios, using realistic mobility traces from TAPASCologne and Luxembourg SUMO Trace.
The table below shows the comparison between the three learning architectures in terms of Model Accuracy and Training Time:
| Method | Accuracy (MSE) | Training Time (s) |
|---|---|---|
| Centralized | 0.0012 | 250 |
| LuST | 0.0016 | 180 |
| MoST | 0.0013 | 210 |
- Centralized Learning: Provides the highest accuracy but incurs higher training time and risks for data privacy.
- Federated Learning: Reduces communication overhead and preserves privacy but sacrifices some accuracy.
- Split Learning: Balances both accuracy and communication overhead, with slightly longer training times.
- Integration of additional learning methods.
- Improvements in energy efficiency and further reduction of CO2 emissions.