Insight247/afrobeat_unsupervised_analysis.py

Unsupervised machine learning analysis of 114,000 Spotify tracks to map where African genres (Afrobeats, Afropop, Highlife) sit in the global sonic landscape, using PCA, K-Means, Hierarchical Clustering, t-SNE, and NMF. # 🌍 Where Does Afrobeats Sit in the Global Music Landscape? ### An Unsupervised Machine Learning Deep Dive into African Music *(Above: A t-SNE projection mapping 114,000 global tracks. The dense cluster of colored points highlights the distinct sonic island of African music against the global gray backdrop.)* --- ## 📌 Project Overview Afrobeats is the biggest music export story of the 21st century. But where does it actually sit in the global sonic landscape — and how does it relate to other African genres, to Latin music, to Hip-Hop? This project answers those questions using unsupervised machine learning — no genre labels, no human-assigned categories. Five complementary ML techniques work together to build a data-driven map of global music, with African genres at the center of the story. > *"The algorithm never saw a single genre label. It found the structure on its own."* ## ❓ Problem Statement Where does Afrobeats actually sit in the global sonic landscape — and how does it relate to other African genres, to Latin music, to Hip-Hop when measured strictly by mathematical audio features? ## 🎯 Problem Specification To fully answer the statement above, this project executes the following objectives: 1. Isolate African genres (e.g., Afrobeat) and contrast them against Global Anchors (Pop, Hip-Hop, Classical) and Diaspora Cousins (Dancehall, Reggaeton, Latin). 2. Determine if African genres are internally diverse or if they share a single monolithic sonic identity. 3. Identify which global genre Afrobeats is most similar to using purely mathematical distances. 4. Extract the underlying, latent "sound archetypes" that make up the African music recipe. --- ## 🗂️ Dataset | Property | Detail | Source | | :--- | :--- | :--- | | **Name** | Spotify Tracks Dataset | Kaggle | | **Author** | Maharshi Pandya | | | **Size** | 114,000 tracks × 20 columns | | | **Genres** | 114 unique genres | | | **License** | Public / Kaggle Community | | ### Audio Features Used | …