🎵 Hit Song Predictor: Apple Music Data Analysis

"Is a hit song made or born?" A machine learning project that analyzes Apple Music metadata and Spotify popularity metrics to predict track success.

🔗 Try the App

👉 Click here to try the Hit Song Simulator

---

📌 Project Overview

What determines the success of a song in the streaming era? This project bridges the gap between Apple Music's rich metadata and Spotify's popularity metrics to perform a Multivariate Regression Analysis. The goal was to quantify the impact of artist fame, momentum, genre, and track duration on a song's performance.

🎯 Key Objectives

1. Data Engineering: Merged Apple Music metadata (Genre, Duration, etc.) with Spotify Popularity scores as a proxy metric.
2. Statistical Analysis: Conducted OLS Regression to identify significant predictors.
3. Predictive Modeling: Built a model with $R^2 = 0.703$, explaining 70% of the variance in popularity.
4. HCI & Application: Developed an interactive "Hit Song Simulator" dashboard for music producers to test "What-If" scenarios.

---

📊 Key Findings

1. The "Formula" for Success ($R^2 = 0.70$)

The analysis revealed that Momentum and Fame are the dominant drivers of success.

• Momentum: A hit song acts as a springboard; high popularity on a previous track significantly boosts the next one (coef = 0.54).
• Fame: Established artist popularity acts as a safety net against failure (coef = 0.42).

2. The "2-Minute Cliff" (Duration Analysis)

By binning song duration into 5 distinct categories, I discovered a non-linear relationship.

• The Cliff (< 2 mins): Songs under 2 minutes suffer a massive penalty (~20 point drop in popularity).
• The Sweet Spot (3:00 - 4:30): This range maximizes the probability of a high score.
• Survivorship Bias (> 6 mins): Very long songs do not crash in popularity, likely because they are mostly released by established artists with dedicated fanbases.

3. Genre & Content Insights

• Cleaned vs. Explicit: Surprisingly, "Cleaned" versions of songs performed best, likely due to wider accessibility (radio play) combined with the virality of the original content.
• Genre Bonus: While most genres performed similarly, Singer/Songwriter tracks received a statistically significant boost (+2.79 points).

---

💻 Features (The Dashboard)

🎛️ Hit Song Simulator

I built a Streamlit web application to make the regression model accessible to non-technical users.

• Interactive Sliders: Users can tweak variables (Fame, Duration, Genre) to see real-time predictions.
• Visual Feedback: The interface provides warnings (e.g., "⚠️ Penalty: Song is too short") based on data insights.
• Batch Analysis: Users can upload a CSV file of demo tracks to process predictions in bulk.

---

🛠️ Tech Stack

• Language: Python 3.9
• Data Analysis: Pandas, NumPy
• Statistics: Statsmodels (OLS Regression), Scipy
• Visualization: Matplotlib, Seaborn
• Web Framework: Streamlit

---

🚀 How to Run Locally

1. Clone the repository

git clone [https://github.com/lucytheboss/apple-music-analysis.git](https://github.com/your-username/apple-music-analysis.git)
cd apple-music-analysis

2. Install dependencies

pip install -r requirements.txt

3. Run the App

streamlit run app.py

---

📝 Limitations & Future Work

• Data Proxy: Spotify popularity was used as a proxy for Apple Music performance due to API limitations. User behavior may differ between platforms.
• External Factors: The model explains 70% of the variance. The remaining 30% is likely driven by unmeasured external factors such as marketing budgets, TikTok trends, and social media buzz.
• Future Work: Incorporating social media sentiment analysis (NLP) could improve the model's accuracy.