Author(s): Bharati Wukkadada
The advent of music streaming platforms in the early 2010s has transformed how music is created, discovered and consumed. These platforms leverage recommendation algorithms that often prioritize popular songs, influencing artists to tailor music for maximum reach. This paper investigates the relationship between song characteristics and popularity, focusing on the Indian market. Data was collected using the Spotify API for the most streamed genres. Exploratory Data Analysis (EDA) provided initial insights into track features and their relationships with song popularity. Songs were categorized into fi e popularity classes, very low, low , medium, high and very high to perform ANOVA test, which revealed significant differences in track features across classes. For machine learning, songs were classified into ‘popular’ (top ~15% with popularity above 65) and ‘unpopular’ categories. Sentiment analysis was conducted, adding a ‘sentiment score’ to the feature set. Various classification algorithms were employed, with logistic regression achieving the highest test accuracy (84.7%), closely followed by other algorithms like support Vector Machines and Random Forests. Key findings revealed that popular songs are generally shorter, exhibit higher instrumentalness, lower speechiness, greater energy, and are louder. Feature importance analysis highlighted song duration as a critical predictor of popularity. This research successfully able to observe the dependence of song popularity on track features and also gain valuable insights from the trends in various audio features.