How Many Genres Are Too Many? Probing Music Classification with AI

The Genre Problem Behind Your Playlist

What makes a song pop, or jazz, or country, or rock? Streaming platforms like Spotify tag songs by genre to help us find new favourites. But what happens when a song fits more than one genre, or none at all? With the increasing popularity of streaming services and their personalized recommendations, automated music genre classification plays a growing role in shaping how listeners discover and experience new music.

A new study by former Bachelor of Computing student Thomas Phan and Dr. Ritu Chaturvedi in the University of Guelph’s School of Computer Science examined how a common machine learning method, the k-nearest neighbours (kNN) algorithm, handles the growing complexity of music genres. Their research shows that as the number of genre categories increases, the algorithm’s accuracy in correctly classifying songs significantly decreases.

Testing AI Limits with Music Categories

To mimic real-world music libraries, the team trained five kNN models using a large Spotify dataset. Each model had an increasing number of genres, from two all the way up to six, including classical, hip-hop, jazz, rock, country and electronic. Each track was described by eight key audio features such as danceability, energy and tempo.

The researchers found a clear pattern: the more genre labels the model had to choose from, the worse it performed. Accuracy dropped from 94 per cent with just two genres to 55 per cent with six. The steepest decline happened when moving from two to three genres, suggesting that even small increases in category count can confuse AI classifiers.

“Music does not fit neatly into boxes,” says Chaturvedi. “As genres evolve and blend, algorithms must adapt, or risk misclassifying what they cannot define.”

Why It Matters for Listeners and Developers

These findings shed light on a growing challenge in music technology: defining genres in a world where artists break traditional boundaries. While genre labels help users explore music, overcomplicating classification can mislead both listeners and recommendation systems.

The study highlights the need for better feature selection and more advanced models that can handle genre overlap. Future work may explore deep learning or hybrid approaches to improve how music streaming platforms understand the songs we love.

Dr. Ritu Chaturvedi, left, and Thomas Phan.

T. Phan and R. Chaturvedi, "Exploring and Mining the Boundaries of Genre Complexity," 2024 Sixth International Conference on Intelligent Computing in Data Sciences (ICDS), Marrakech, Morocco, 2024, pp. 1-7, doi: 10.1109/ICDS62089.2024.10756310.

This story was written by Mojtaba Safdari as part of the Science Communicators: Research @ CEPS initiative. Mojtaba is a PhD candidate in the School of Engineering under Dr. Amir A. Aliabadi.