🎧 Spotify Tracks EDA — Exploratory Data Analysis

A detailed exploratory data analysis on Spotify track data to uncover musical trends, popularity patterns, and the characteristics that make songs successful.

---

📘 Project Overview

This project focuses on Exploratory Data Analysis (EDA) of Spotify track data. It aims to identify relationships between audio features (like energy, valence, danceability) and a song’s popularity, uncovering trends across decades and genres.

The study reveals how modern music differs from older eras — becoming shorter, more energetic, less acoustic, and more dance-focused — while offering insights useful for music producers, analysts, and data scientists.

---

🎯 Objectives

The primary goals of this analysis were:

• To understand the distribution of key musical features (popularity, energy, danceability, valence, etc.)
• To explore relationships between popularity and musical characteristics.
• To examine temporal trends (year-wise analysis).
• To identify top-performing tracks and artists.
• To derive recommendations for model design or music success strategies.

---

📊 Dataset Summary

Feature	Description
track_id	Unique Spotify track identifier
track_name	Song title
artist_name	Artist name
year	Year of release
popularity	Popularity score (0–100)
danceability	Suitability for dancing
energy	Intensity of the song
valence	Positiveness or musical mood
acousticness	Degree of acoustic elements
instrumentalness	Likelihood of instrumental-only
speechiness	Spoken word presence
tempo	Beats per minute (BPM)
duration_ms	Song length in milliseconds
language	Detected lyric language
loudness, mode, key, time_signature	Musical composition features

---

🔍 Univariate Analysis

• Popularity → Highly skewed; most songs have low scores, few are global hits.
• Danceability → Left-skewed; majority are rhythmic and danceable.
• Energy → Concentrated near 1.0; most tracks are high-energy.
• Valence → Bimodal; both happy and neutral/sad songs exist widely.
• Tempo → Bimodal peaks around 120 and 140 BPM.
• Duration → Mostly 3–5 mins; extreme outliers exist.
• Loudness → Unreliable feature due to data errors.
• Instrumentalness → Dominated by vocal tracks (≈0).

---

🔗 Bivariate Analysis

• Popularity vs Danceability → Popular songs have Danceability > 0.4.
• Popularity vs Energy → Low Energy = low popularity; high Energy ≠ guaranteed success.
• Popularity vs Duration → Negative correlation; short songs perform better.
• Popularity vs Valence → Mood doesn’t predict hits.
• Energy vs Loudness → Highly correlated; redundant.
• Danceability vs Tempo → Weak relationship; tempo not a predictor.
• Energy vs Popularity Category → High/Medium hits have higher energy.
• Danceability vs Popularity Category → Hits more danceable and consistent.
• Valence vs Popularity Category → Mood uniform across categories.

---

🧠 Correlation Insights

Feature	Correlation with Popularity
Energy	0.15
Valence	0.15
Danceability	0.14
Acousticness	-0.14
Instrumentalness	-0.13

✅ Recommendation: Drop redundant pairs — Energy-Loudness (0.90), Valence-Danceability (0.64).

---

📅 Temporal Trends

• Energy ↑ — Music became more intense since 1970s.
• Valence ↓ — Songs became less positive post-1995.
• Danceability ↑ — Spike after Disco era (1980s).
• Speechiness ↑ (2020s) — Rise in rap/spoken content.
• Acousticness ↓ — Shift to electronic music post-1990s.
• Duration ↓ — Songs shorter in streaming age.
• Popularity Distribution — >80% of tracks are low popularity.

---

🌍 Language & Artist Analysis

• Korean tracks have highest average popularity (~27.5).
• Hindi ranks next (~18.5); Malayalam lowest (~7).
• Artists by Volume: Shankar Mahadevan, Ramin Djawadi, Alan Silvestri (film composers).
• Artists by Popularity: Rihanna, Taylor Swift, Kendrick Lamar, LE SSERAFIM (mainstream hits).

---

💡 Key Takeaways

1. ✅ Use Classification Models: Predict “Hit” or “Miss” instead of regression on scores.
2. 🧹 Data Cleaning: Remove soundtrack composers and outliers.
3. ⚡ Apply Feature Filters: Exclude songs with very low energy/danceability.
4. 🎭 Combine Mood Features: Use interaction of energy × valence.
5. 🔁 Feature Reduction: Drop redundant variables (e.g., Loudness).

---

⚙️ Tech Stack

• Language: Python 3.11
• Libraries: Pandas, NumPy, Matplotlib, Seaborn, Plotly
• Environment: Jupyter Notebook

---

🚀 Run Instructions

git clone https://github.com/yourusername/spotify-eda.git
cd spotify-eda
pip install -r requirements.txt
jupyter notebook

---

📈 Future Work

• Build classification model for hit prediction.
• Add time-series trend forecasting.
• Sentiment analysis using lyrics.
• Create interactive dashboard (Streamlit/Dash).

---

👤 Author

MANISH KUMAR ROY