🧠 Unsupervised Machine Learning Portfolio

A collection of unsupervised machine learning projects built with Python and Scikit-learn, exploring clustering, dimensionality reduction, probabilistic modelling, and recommendation systems on real-world datasets.

---

📁 Projects Overview

Folder	Technique	Description
kmean-hierarchical_clustering	K-Means + Hierarchical	Segment data into meaningful groups
gaussian_mixture	Gaussian Mixture Models	Probabilistic soft-assignment clustering
principal_component_analysis	PCA	Reduce high-dimensional data while preserving variance
dimensional_reduction	t-SNE / UMAP	Non-linear dimensionality reduction for visualisation
recommendation_system	Collaborative Filtering	Build a personalised item recommendation engine
clustering_geospatial	K-Means + GeoPandas	Cluster world countries by GDP and population on a map

---

🗂️ Project Details

1. 📍 Clustering & Geospatial Analysis

Applied K-Means clustering on world country data using GDP and population as features, then visualised the results as choropleth maps and multidimensional scatter overlays.

Key techniques:

• K-Means clustering on economic indicators
• Geospatial visualisation with GeoPandas
• Multidimensional plotting — colour (GDP), size (population), shape (continent)
• Coordinate Reference System (CRS) handling

Dataset: Natural Earth Low Resolution — country geometries, GDP, population, continent

---

2. 📊 K-Means & Hierarchical Clustering

Compared flat (K-Means) and hierarchical (agglomerative) clustering approaches on structured datasets. Used the elbow method to determine optimal cluster count and dendrograms to visualise merge distances.

Key techniques:

• K-Means with elbow method for optimal k
• Agglomerative hierarchical clustering
• Dendrogram visualisation
• Silhouette score evaluation

---

3. 🌊 Gaussian Mixture Models

Modelled data distributions using probabilistic Gaussian Mixture Models (GMM), allowing soft cluster assignments where each data point has a probability of belonging to each cluster.

Key techniques:

• Expectation-Maximisation (EM) algorithm
• Soft vs hard cluster assignment
• BIC/AIC model selection
• Comparison with K-Means hard clustering

---

4. 📉 Principal Component Analysis (PCA)

Applied PCA to reduce high-dimensional datasets to 2–3 components while retaining maximum variance, enabling faster modelling and cleaner visualisations.

Key techniques:

• Explained variance ratio analysis
• Scree plots for component selection
• 2D and 3D PCA projections
• PCA as preprocessing for downstream ML

---

5. 🔍 Dimensionality Reduction (t-SNE / UMAP)

Used non-linear dimensionality reduction techniques to reveal hidden structure in high-dimensional data that PCA cannot capture linearly.

Key techniques:

• t-SNE for cluster visualisation
• UMAP for faster non-linear projection
• Comparison of linear vs non-linear reduction
• Hyperparameter tuning (perplexity, n_neighbors)

---

6. 🎯 Recommendation System

Built a collaborative filtering recommendation engine that suggests items to users based on patterns in historical interactions — the same approach used by Netflix and Spotify.

Key techniques:

• User-based and item-based collaborative filtering
• Cosine similarity for user/item matching
• Matrix factorisation concepts
• Evaluation with precision and recall

---

🛠️ Tech Stack

Library	Purpose
scikit-learn	Clustering, PCA, GMM, model evaluation
pandas	Data loading, manipulation
numpy	Numerical computation
matplotlib	All visualisations and plots
seaborn	Statistical visualisation
geopandas	Geospatial data handling and mapping
shapely	Geometric operations

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/Lami14/alx_data_science.git
cd alx_data_science

2. Install dependencies

pip install scikit-learn pandas numpy matplotlib seaborn geopandas shapely

3. Open any project

cd kmean-hierarchical_clustering
jupyter notebook

Each folder contains its own notebook with inline explanations and outputs.

---

💡 What I Learned

• How unsupervised learning finds hidden structure in unlabelled data
• When to choose K-Means vs GMM vs hierarchical clustering for a given problem
• How PCA and t-SNE complement each other — PCA for speed, t-SNE for visualisation
• How real-world recommendation systems work under the hood using similarity metrics
• How to represent economic and demographic data visually on geographic maps

---

🔮 Future Improvements

• Add evaluation metrics (silhouette score, Davies-Bouldin index) across all projects
• Deploy the recommendation system as a Flask API
• Add an interactive map dashboard using Folium or Plotly
• Extend geospatial analysis with South African provincial data

---

📄 License

MIT License — feel free to fork and build on this work.

---

Built by Lamla · Unsupervised ML Portfolio · ALX Africa Data Science Programme 🇿🇦