leks39/Naija-Songs-into-Genres

Clustering Nigerian songs into genres using K-Means # Naija-Songs-into-Genres Clustering Nigerian songs into genres using K-Means This exercise uses an unsupervised learning algorithm known as K-Means clustering combined with a dimensionality reduction technique PCA to group Nigerian songs into sub-genres from the 'Afrobeats' main genre. The dataset is made up of 500+ songs released over a 20-year period (2000–2020). Audio Features such as; danceability, acoustiness, energy, instrumentalness, liveness, loudness, speechiness, tempo, and popularity were used for clustering. Audio features are defined as a range of quantitative metrics that create a relatable and relevant profile for a song. While a genre is defined as a group of songs that shares a distinct style, form, pattern, and sometimes origin. Check out the definition of these features on Spotify The aim of this study was to separate Nigerian songs into distinct groups using audio features instead of lumping all songs under the Afrobeats umbrella. 503 songs were segmented into 5 clusters using 6 principal components which explained approximately 76% of the variation in the dataset. No surprise Afropop and Afro Dancehall are the most popular genres given the importance Nigerian musicians place on danceability and energy as audio features. Alte is a new wave sound, had the lowest no of entries. The most difficult genre to name was Afro-Fusion, this was because the genre contained artists that blended multiple genres like; Alte, Highlife, and RnB to create one sound.