This project applies unsupervised learning techniques to analyze global child mortality data from 1990 to 2019, sourced from Our World in Data. The goal is to uncover patterns in causes of death among children under five and cluster countries based on similar mortality trends using techniques like PCA, SVD, K-Means, and Hierarchical Clustering.
- Identify key causes of death across countries and years.
- Use dimensionality reduction (PCA, SVD) to simplify high-dimensional data.
- Apply clustering algorithms to uncover country groupings with similar mortality patterns.
- Support global health policy by extracting interpretable insights.
- Data Source: Our World in Data – Causes of death in children under 5 (1990–2019)
- Preprocessing:
- Removed missing and cumulative 'World' entries
- Renamed lengthy columns for clarity
- Aggregated data across years per country
- Dimensionality Reduction:
- PCA and SVD used to extract 7 principal components explaining 90%+ variance
- Clustering:
- K-Means: Optimal clusters determined via Elbow Method
- Hierarchical Clustering: Dendrograms plotted with complete and centroid linkage
- Top causes of death: respiratory infections, preterm birth, and diarrheal diseases
- Countries with the highest death tolls: India, Nigeria, Pakistan, Ethiopia
- PCA revealed 7 components capturing most variance
- K-Means with
k=29and Hierarchical Clustering exposed meaningful global groupings by region and mortality pattern
- Scree plots for PCA and SVD
- Choropleth map of total deaths by country and year
- Cluster plots from K-Means and hierarchical dendrograms
- Trends over time for top causes of death
- Python: Pandas, NumPy, Matplotlib, Seaborn, Plotly, Scikit-learn
- Machine Learning: PCA, SVD, KMeans, Agglomerative Clustering