Event Type

Research Presentation

Academic Department

Mathematics and Statistics

Location

Dana Science Building, 2nd floor

Start Date

25-4-2025 1:00 PM

End Date

25-4-2025 2:30 PM

Description

Under the direction of Dr. Giancarlo Schrementi

Music genres are typically categorized based on auditory characteristics, but some genres, such as sleep, chill, ambient, and study music, share striking similarities. This paper aims to provide insights into the mathematical differences underlying genre distinctions. Initial exploratory data analysis is performed on a dataset from Spo- tify that contains numeric measures for tracks across these four genres and statistical differences are noted. Three classification models, a discriminative model (Logistic Re- gression) and two generative models (Linear Discriminant Analysis and Naive Bayes) are trained and then used to predict the genre of novel tracks. All the models are able to distinguish the genres, but have different patterns of error. These modeling results demonstrate that there is a measurable mathematical difference between the genres.

Share

COinS
 
Apr 25th, 1:00 PM Apr 25th, 2:30 PM

Discriminative VS Generative Classification Models : Classify Music Genres In A Spotify Dataset

Dana Science Building, 2nd floor

Under the direction of Dr. Giancarlo Schrementi

Music genres are typically categorized based on auditory characteristics, but some genres, such as sleep, chill, ambient, and study music, share striking similarities. This paper aims to provide insights into the mathematical differences underlying genre distinctions. Initial exploratory data analysis is performed on a dataset from Spo- tify that contains numeric measures for tracks across these four genres and statistical differences are noted. Three classification models, a discriminative model (Logistic Re- gression) and two generative models (Linear Discriminant Analysis and Naive Bayes) are trained and then used to predict the genre of novel tracks. All the models are able to distinguish the genres, but have different patterns of error. These modeling results demonstrate that there is a measurable mathematical difference between the genres.