Music Data Processing using Spotify and AI

This project creates a Spotify data analysis which analyses track data to discover patterns in audio features, popularity, and genres. This is basically for audiophiles.

Comment

Inspiration

Music connects with everyone, but we wanted to dig into why certain songs just fit specific moods. As Spotify users, we kept noticing these patterns—like, upbeat songs for workouts, calm ones for relaxing, danceable stuff for parties. But what actually makes a song work for those moments? And do genres like "Rock" or "EDM" really sound that different, or is it just labels? We figured, why not use data to crack this? Turned into a mission: connect audio features, genres, and what listeners like into something useful for everyone.

What It Does

Basically, our project breaks down Spotify tracks to answer stuff like:

  • How loudness, energy, and danceability are linked (spoiler: loud = energetic).
  • What songs are perfect for parties (hint: high energy + danceable = winner).
  • Can we group songs by their audio "DNA"? Turns out, yes.
  • How genres like EDM vs. Acoustic actually measure different.

The dashboard? It shows trends—top artists/genres, how energy ties to "happy" vibes, why today’s hits are all loud and energetic. Simple but kinda eye-opening.

How We Built It

  1. Data Pipeline: Pulled track data (features, genres, popularity) from Spotify’s API, cleaned it up using Python—Pandas did most the heavy lifting.
  2. Visualizations: Made heatmaps for correlations, PCA plots for genres, all with Seaborn/Matplotlib.
  3. Clustering: Used K-means (elbow method said 4-5 clusters) to group songs, then PCA to visualize.
  4. Dashboard: Built it with Plotly and Tableau—interactive charts for stuff like popularity, tempo ranges, loudness-energy trends.

Challenges We Ran Into

  • Genre Chaos: So many tracks had weird genre mixes, like "Indie-Folk-Pop-Rock." Took ages to sort them.
  • Cluster Confusion: Math said 4-5 clusters, but naming them? Like, is this group "party" or "study"? Had to guess.
  • Slow Dashboards: Scatterplots with 10k+ points lagged like crazy. Took a lot tweaking to fix.

Accomplishments We’re Proud Of

  • Loudness = Energy: Proved they’re super linked (~0.8!), and acoustic tracks? Neither loud nor energetic.
  • Guess the Genre: Predicted EDM/Acoustic with 85% accuracy using just 3-4 features. Not bad!
  • Hidden Party Gems: Found 200+ underrated tracks that balance danceability and energy—DJs gonna love these.

What We Learned

  • Pop’s Loudness War: Modern hits are loud and energetic for a reason—they’re designed to stick.
  • Clustering Magic: Squishing 10+ features into 2D clusters actually worked. Calm vs. hype songs? Totally separate.
  • Genres Aren’t Random: EDM is danceable, Rock’s energy is all over the place, Acoustic’s minimalism is measurable.

What’s Next

  • Mood-Based Picks: Add lyrics analysis to recommend songs based on your vibe. Feeling sad? Here’s a playlist.
  • Smarter Playlists: Auto-create playlists that switch from gym beats to chill tunes without you lifting a finger.
  • Tool for Artists: Help musicians see gaps—like "your track needs more bass to compete in pop."
  • Predict Viral Songs: Mix Spotify data with TikTok trends to guess the next big hit. Because algorithms rule now.

By blending data and music,it makes discovering and creating music better—one scatterplot at a time.

Built With

Share this project:

Updates