Business Problem Definition
The music industry has undergone significant changes, transitioning from physical albums and CDs to digital streaming platforms. This shift has not only transformed how music is shared but also broadened the range of popular styles, creating a fast-evolving and diverse market where genres often blend. Modern hit songs now incorporate elements from various genres like pop, electronic, and R&B, with changes in tempo, energy, and melody to cater to audience preferences.
With advancements in data and AI, music companies and artists use audio feature analysis to predict what makes a song popular. Our project aims to evaluate the popularity across 14,034 music tracks from Free Music Archive (FMA), uncovering the key factors to high listen counts.
Motivation
As the music industry and streaming services have become increasingly competitive, it is important to understand what attracts users to listen frequently. By evaluating the correlation between music features and listening counts, we can discover elements that encourage users’ engagement. This insight can further help the music industry and streaming platforms to enhance their popularity.
Report Summary
This report presents an exploratory analysis of music track popularity using data from the Free Music Archive (FMA). We begin by detailing the data cleaning and merging process for the three main datasets: tracks, audio features, and genres. Next, we use data visualizations to give an overview of the dataset, followed by an analysis of the correlation between audio features and the number of listens. Finally, we summarize our findings, discuss challenges encountered, suggest directions for future analysis, and conclude with references.
Key findings include:
Pure music (tracks without lyrics) generally received higher listens compared to tracks with vocals, indicating a strong audience preference for instrumental tracks. Track duration was found to have no strong correlation with popularity, suggesting that both short and long tracks can perform well. Genre distribution and track characteristics were visualized, revealing distinct patterns between different genres, such as high danceability for Electronic and Rock genres. Analysis of audio features indicated no clear linear correlation with track listens, pointing to the need for more sophisticated regression analysis using machine learning techniques. Social factors, such as artist recognition and song popularity, were positively associated with track listens, emphasizing the importance of social engagement in driving track success.
Project contributors:
Lifu (Leaf) Li
Lujia (Lucia) Yu
Emre Can Baykurt
Pin-Hao (Johnny) Pan
Kuang-Ching (Amanda) Ting