Skip to content

danilogiarlini/Global-CO2-Emission-Forecasting

 
 

Repository files navigation

Global CO2 emission forecasting

Project Overview

Team

Ajay Krishna & Danilo Giarlini

Source

"Emission by Country" from Kaggle: https://www.kaggle.com/datasets/thedevastator/global-fossil-co2-emissions-by-country-2002-2022

Dataset structure

11 columns and 63000 rows (reduced to around 3000 for our purpose). Columns are: time, total emission, emission from coal, emission from gas etc.

Problem type

Multivariate Time series

Framework used

Vector Autoregression (VAR)

Goal

Design a multivariable time-series model that predicts the total and emissions per energy sector for a selected number of countries.

(Preliminary) Motivation step

Visualize the total emission based on input features (using Python or Tableau).

Problem statement

Pick 10 categories (3 developed countries, 3 developing countries, 3 under developed countries and 1 global).

  1. Design a multivariable time-series model.
  2. Use VAR to predict the total and emissions per energy sector.

Approximate timeline of the project

Day 1

  1. Explore the data.
  2. Select the data for 10 categories mentioned above.

Day 2

  1. Check if the dataset satisfies the prerequired tests of VAR (Granger’s Causality Test and Cointegration test). If the tests are failing, try dropping some columns or try trimming the data (as an example). And then, run the tests again.
  2. Implement VAR in python for one country (one of the subdatasets).

Day 3

  1. Write a for loop and check if the VAR approach works for all the selected countries.
  2. Plot the results.
  3. Compare the accuracy in prediction (for the test set) for different countries.

Day 4

  1. Prediction using VAR for unseen time point, i.e., for the data set, predict the emissions from 2023 to 2030.

  2. Plot the prediction.

  3. Check if a mathematical relation between the total emissions and the sector-based emissions can be extracted from VAR. If yes, then

    3a. evaluate which sector contributes the most to the total emission and

    3b. think of a way whether sectors can be prioritized (e.g., feature importance concept using Lasso, Ridge).

Day 5-7

Further improvements and learning.

Day 7

Presentation of results and insights.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 100.0%