- Project Goal: Identify which audio characteristics influence the popularity of tracks on Spotify.
- Dataset: 114,000 songs from the Spotify Tracks dataset on Kaggle.
- Approach: Conduct exploratory data analysis (EDA) and build a basic linear regression model using two features:
danceabilityandenergy.
Background & Problem
2 Data Loading & Setup
library(readxl)
dataset <- read_excel("~/Downloads/dataset.xlsm")
## New names: ## • `` -> `...1`
suppressPackageStartupMessages({
library(ggplot2)
library(readxl)
library(tidyverse) # dplyr, ggplot2, etc.
library(caret) # machine-learning helpers
library(GGally)
library(conflicted)
})
conflict_prefer("filter", "dplyr")
conflict_prefer("lag", "dplyr")
3 Data Cleaning & Wrangling
library(tidyverse)
df <- dataset %>%
select(-any_of(c("...1","track_id","artists",
"album_name","track_name","track_genre"))) %>%
mutate(across(where(is.logical), as.numeric)) %>%
drop_na()
glimpse(df)
## Rows: 114,000 ## Columns: 15 ## $ popularity <dbl> 73, 55, 57, 71, 82, 58, 74, 80, 74, 56, 74, 69, 52, 6… ## $ duration_ms <dbl> 230666, 149610, 210826, 201933, 198853, 214240, 22940… ## $ explicit <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,… ## $ danceability <dbl> 0.676, 0.420, 0.438, 0.266, 0.618, 0.688, 0.407, 0.70… ## $ energy <dbl> 0.4610, 0.1660, 0.3590, 0.0596, 0.4430, 0.4810, 0.147… ## $ key <dbl> 1, 1, 0, 0, 2, 6, 2, 11, 0, 1, 8, 4, 7, 3, 2, 4, 2, 1… ## $ loudness <dbl> -6.746, -17.235, -9.734, -18.515, -9.681, -8.807, -8.… ## $ mode <dbl> 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0,… ## $ speechiness <dbl> 0.1430, 0.0763, 0.0557, 0.0363, 0.0526, 0.1050, 0.035… ## $ acousticness <dbl> 0.0322, 0.9240, 0.2100, 0.9050, 0.4690, 0.2890, 0.857… ## $ instrumentalness <dbl> 1.01e-06, 5.56e-06, 0.00e+00, 7.07e-05, 0.00e+00, 0.0… ## $ liveness <dbl> 0.3580, 0.1010, 0.1170, 0.1320, 0.0829, 0.1890, 0.091… ## $ valence <dbl> 0.7150, 0.2670, 0.1200, 0.1430, 0.1670, 0.6660, 0.076… ## $ tempo <dbl> 87.917, 77.489, 76.332, 181.740, 119.949, 98.017, 141… ## $ time_signature <dbl> 4, 4, 4, 3, 4, 4, 3, 4, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4,…
Track Popularity Distribution
ggplot(df, aes(popularity)) + geom_histogram(binwidth = 3, fill = "steelblue") + labs(title = "Distribution of Track Popularity", x = "Popularity (0 – 100)", y = "Count")
Popularity vs. Explicit Content
Energy Distribution
Data Splitting for Modeling
library(caret) set.seed(123) idx <- createDataPartition(df$popularity, p = 0.7, list = FALSE) train <- df[idx, ] test <- df[-idx, ]
Linear Model Output
| Term | Estimate | SE | t-value | p |
|---|---|---|---|---|
| (Intercept) | 30.501 | 0.319 | 95.716 | 0.000 |
| danceability | 4.900 | 0.460 | 10.643 | 0.000 |
| energy | -0.055 | 0.317 | -0.175 | 0.861 |
Model Performance Summary
| R² | Adj R² | σ (resid SE) | F-test p |
|---|---|---|---|
| 0.001 | 0.001 | 22.304 | 0 |
## ## **Test-set RMSE:** 22.26
Final Evaluation Metrics
pred <- predict(model, test)
rmse <- sqrt(mean((pred - test$popularity)^2))
r2 <- cor(pred, test$popularity)^2
cat("RMSE:", round(rmse, 2), "\n")
## RMSE: 22.26
cat("R-squared:", round(r2, 2), "\n")
## R-squared: 0
Conclusion & Takeaways
- Danceability contributes positively to track popularity.
- Energy did not significantly predict popularity in our model.
- Spotify popularity is likely influenced by external factors not captured here, like artist fame, marketing, and playlist placement.
- This project helped apply real-world EDA, modeling, and communication skills using R.