Cities of Tomorrow – Urban Sustainability Analysis
A Data-Driven Exploration of Urban Sustainability Metrics Using Advanced Analytics & Machine Learning.
Project Overview
Urban sustainability has become a critical challenge as cities grow rapidly and environmental pressures increase. This project provides a comprehensive end-to-end data science workflow to analyze, visualize, and model sustainability indicators across global cities.
The goal is to explore questions such as -
Which cities are leading in sustainability?
How do green cover, renewable energy usage, and air quality influence sustainability scores?
Can we cluster cities based on their sustainability profiles?
Which factors contribute most to sustainability outcomes?
Dataset
The dataset contains multiple sustainability-related features such as -
🌿Green Cover Percentage.
🏭Air Quality Index.
🔋Renewable Energy Usage.
🚇Public Transport Efficiency.
🏘Population Density.
🌞Solar & Wind Energy Capacity.
🌍Urban Sustainability Score (Target variable).
Exploratory Data Analysis (EDA)
✔ Correlation Heatmap - Analyzes relationships among numeric variables using seaborn.
✔ Fabric-Themed KDE Plots - Beautiful density distribution plots with custom palettes.
✔ Pairplot (KDE-based) - Reveals variable relationships & distributions.
✔ Outlier Analysis - Boxplots detecting variability across sustainability indicators.
✔ PCA (Dimensionality Reduction) - Maps high-dimensional data into 2D principal components for better visualization.
✔ Hierarchical Clustering & Dendrogram - Provides insights into natural clusters among cities.
Machine Learning Modeling
This project incorporates multiple machine learning techniques to predict Urban Sustainability Scores using diverse city-level environmental, infrastructural, and socio-economic features. The modeling step helps quantify how different factors contribute to sustainability outcomes across global cities.
Models Used
1) XGBoost Regression - XGBoost is used as the primary model thanks to its -
High performance on structured/tabular data.
Ability to handle non-linear relationships.
Built-in regularization to prevent overfitting.
It predicts the Urban Sustainability Score based on variables such as -
Green cover percentage.
Air quality index.
Waste recycling efficiency.
Renewable energy adoption.
Urban mobility efficiency, etc.
2) Random Forest Regression - Random Forest serves as a secondary model for comparison. It provides -
Strong generalization.
Robustness to noise.
Reliable feature-importance insights.
Using two models enables a deeper understanding of variable influence and ensures consistency of predictions.
Model Evaluation Metrics
MAE
RMSE
R² Score
Residual Diagnostics
Residual analysis was performed to check for model bias, heteroscedasticity, and distributional behavior.
The following diagnostic plots were generated -
✅Residual Distribution
Shows whether residuals follow a normal distribution.
Helps identify skewness or model bias.
Highlights outliers that the model struggles to predict.
✅Residuals vs. Predictions Plot
Evaluates homoscedasticity (constant variance).
Ideally, residuals should scatter randomly around zero.
Patterns in this plot suggest missing features or non-linearity.
These diagnostics verify that the model is reliable and generalizes well.
Feature Importance Analysis
Both models provide insights into which factors most affect urban sustainability -
Renewable energy usage.
Pollution control metrics.
Public transport efficiency.
Waste management quality.
Green cover.
This analysis helps policymakers prioritize impactful interventions.
Clustering & PCA Analysis
Cities are grouped using -
1) PCA → 2D Projection - Reduces dimensions for better cluster visualization.
2) KMeans Clustering - Groups cities into meaningful segments based on sustainability metrics.
Technologies Used
| Languages | Python |
| Visualization | Matplotlib, Seaborn, Plotly |
| Data Processing | Pandas, NumPy |
| Machine Learning | Scikit-Learn, XGBoost |
| Clustering | KMeans, Hierarchical Clustering |
| Dimensionality Reduction | PCA |
| Environment | Kaggle Notebook |
Key Insights
1) Cities with higher renewable energy usage significantly outperform others in sustainability scores.
2) Air quality and green cover are the two most dominant factors.
3) PCA & clustering revealed 3 distinct types of sustainable cities -
Renewable-heavy cities.
High green cover but low air quality.
Balanced mid-tier sustainability cities.
Conclusion
This project demonstrates the power of data-driven sustainability analysis, combining -
Strong EDA.
Rich interactive visualizations.
Dimensionality reduction.
Clustering.
Predictive modeling.
The results provide a holistic understanding of what makes a city sustainable and how future urban spaces can be designed.
