Carbon emission Analysis & Prediction

A project by the AIGLass Team - Explore Data Science Academy Internship Project

Project Developers

Israel Ezema
Ahmad Barde
Damilare
Ebere
Raphael
Teddy Waweru

Project Description

Since the discoveries around the effects of increased carbon emissions on the climate, multiple initiatives have been developed to attempt to curb the issue.

This machine learning project attempts to offer a data-driven solution for the same agenda. Moreover, the project intends to offer predictive capabilities utilizing data analytics & machine learning models.

This is only intended as an overview of the work that was carried out. Find project links at the bottom for the accompanying Research paper, Jupyter Notebooks, Presentation Slides & Application Deployment

Expectations

The project specifically aims at complementing the implementation of carbon credits in reducing carbon emissions.

The general idea of carbon credits is that organizations, institutions & countries are able to purchase in carbon credits as a form of offsetting their overall carbon emissions. These credits are tied to projects that actively remove carbon from the environment, such as building renewable energy plants, planting trees, protecting forest reserves.

Where does this project lie in the carbon credit transactions? The solution enables carbon credit purchasers to develop emission predictions. This offers a baseline insight to complement long-term planning of the purchaser’s expenses & ton organizing prospective projects. The predictions may have a range in error, but this is only due to improve with time & reconsderation of affecting factors.

Data sources

The data utilized in the project are certified under GPL.

Data description/breakdown

Project Strategy

Data analysis

The bulk of the work entailed deriving analytical benefits from the data that was sourced, & understanding how the final model would be affected.

Some of the features collected showed little benefit to the,

Machine Learning model

The final model that was utilized in the web application was the XGBoost, with an RMSE of 15.4. This model performed best in comparison to linear & tree based algorithms that were trained: (decision-trees, RandomForest, Ridge & Lasso Regression).

Future considerations shall be placed on neural networks. However, it is anticipated that the con of runtime for the model will outweigh the benefit of improved model predictions.

Conclusion

The project successfully highlighted the prominent features that are causal to carbon emission according to the dataset that was utilised. The model’s error range is feasible for the purpose of the solution, but there are opportunities to improve it.

Heightened carbon emissions on earth is asuredly a future generation’s problem, and some of the effects already being experienced currently via climate change. The best that our society can offer is by utilising the tools that we have to progress forward; hoping that we shall finally find the silver bullet that will end the worry. Until then, we continue the toil.

Associating Links

Web Application hosted on Heroku Platform.

Note: The application may take some time to load if at all it has not been accessed for some time. Give it some time to ‘wake’, else reload the browser tab

Final Presentation to the EDSA Internship supervisors

Research Paper **to be availed soon