← Back to Projects Time-Series Project

CO₂ Emission Analysis & Forecasting

A time-series analysis and forecasting project that models global CO₂ emissions using machine learning, feature engineering, and compares recursive and direct prediction strategies.

Python Time Series Machine Learning Regression Forecasting

Problem And Solution

Problem

Understanding CO₂ emission trends requires not only historical analysis but also reliable forecasting to anticipate future environmental impact.

Solution

Built a system that combines exploratory analysis with machine learning-based forecasting to analyze past trends and predict future emissions.

Data And Feature Engineering

Dataset

Global CO₂ emission dataset (Kaggle) covering historical emissions from ~1700 to 2023.

Feature Engineering

  • Lag features (lag1, lag2)
  • Growth rate calculation
  • Year as temporal feature

Exploratory Data Analysis

  • Rapid growth during industrialization (1950–1970)
  • Acceleration post-2000
  • Recent stabilization trends due to global initiatives

Model Development

Baseline

Lag-based baseline model (MAE ≈ 0.35)

Models Used

  • Linear Regression
  • Random Forest Regression

Insights

Random Forest captured non-linear patterns better, while Linear Regression provided stable trend estimation.

Forecasting Techniques

Recursive Forecasting

Uses previous predictions as input. More realistic but error accumulates over time.

Direct Forecasting

Predicts future values independently. More stable but less adaptive.

Results And Insights

Forecasts indicate slower emission growth and potential stabilization trends, reflecting global sustainability efforts and policy changes.

Future Improvements

  • ARIMA / SARIMA models
  • Tableau dashboard integration
  • Country-level forecasting