Overview

This Data Science project was based on a dataset from Spotify (available on Kaggle)  with the attributes of 160,000+ Spotify tracks

(https://www.kaggle.com/yamaerenay/spotify-dataset-19212020-160k-tracks).

Variables

The dataset contains multiple attributes  like acousticness, danceability, energy, explicit, instrumentalness, key, liveness, loudness, mode, speechiness, tempo, valence, duration, artists, popularity, year for each track.

 These were used to construct a Predictive Model to predict the Popularity of a song based on its  attributes using Linear Regression model.

Data Cleaning

spotify.csv

• Some tracks had the same names and artists but had slightly varying attributes. These duplicates were  removed.
• After removing all duplicate tracks, 155,000 tracks remained.

 Key Results

Two Regression models were created .

Model 1 (popularity ~ year): Predicting Popularity based on year

• The year a track was released explains about 77% of the variance in its popularity.
• The later a track is released by a year, the more its average popularity by about 0.74 points.

Model 2 (popularity ~ key + liveness + loudness + tempo + speechiness + instrumentalness + acousticness + danceability + explicit + valence)

• The model explains about 45% of the variance in the popularity of a track.
• The reference key and explicit in this model are both 0.
• Tracks in certain keys tend to be less popular compared to tracks in the 0 key.
• The more liveness, instrumentalness, and speechiness, acousticness, and valence a track has, the more it loses its popularity on average.
• If a track has explicit lyrics, its popularity will rise by about 10 points on average.
• The louder the track, the higher its popularity is on average.
• The more a track’s danceability, the higher its popularity is on average.

Business Applications

• While it may be difficult to program a song to be highly popular, the model can be used by record labels to determine which album will sell based on the attributes of each track in the album. This allows record labels to determine how to allocate resources for each album.