🖥️ Live Dashboard

🚀 Open the interactive Streamlit dashboard:
👉 streaming-user-behavior-eda-fpxbmrvygfyrcqvpflpykj.streamlit.app

Explore content performance, user segmentation, retention signals, and growth opportunities through a 4-page interactive dashboard built with Streamlit and Plotly.

📊 Streaming Platform User Behavior Analysis

A Data Story on Content Consumption, Retention, and Growth Opportunities

An end-to-end exploratory data analysis project that uncovers actionable insights from streaming platform user behavior — simulating the kind of data-driven decision making used by growth and product teams at audio/video streaming platforms like NOICE.

📌 Problem Statement

Understanding how, when, and why users engage with content is critical for any streaming platform looking to grow. This project takes a storytelling-first approach to EDA — going beyond charts and statistics to deliver business insights that product, growth, and content teams can act on.

Key questions explored:

• What content formats and genres drive the highest engagement?
• When do users drop off, and what predicts early churn?
• Which user segments are most valuable — and most at risk?
• What does the data suggest about content strategy going forward?

🎯 Objectives

• Perform comprehensive EDA on simulated streaming platform interaction data
• Translate raw patterns into clear, business-relevant narratives
• Identify retention risks and growth opportunities from behavioral signals
• Present findings in a structured, stakeholder-friendly format

🗂️ Project Structure

streaming-eda-data-story/
├── data/
│   ├── raw/                            # Raw simulated datasets
│   ├── processed/                      # Cleaned & merged datasets
│   └── generate_synthetic_data.py      # Reproducible dataset generator
├── notebooks/
│   ├── 01_data_overview.ipynb          # Dataset structure & quality check
│   ├── 02_content_analysis.ipynb       # Content consumption patterns
│   ├── 03_user_segmentation.ipynb      # User behavior segmentation
│   ├── 04_retention_analysis.ipynb     # Drop-off & churn signals
│   ├── 05_growth_opportunities.ipynb   # Actionable growth findings
│   └── 06_executive_summary.ipynb      # Full data story (end-to-end)
├── visuals/                            # Exported charts & plots
├── report/                             # Stakeholder-ready summary report
├── requirements.txt
├── .gitignore
└── README.md

📸 Dashboard Preview

Preview snapshots from the full-scale synthetic dataset: 2,000 users · 500 content items · 50,000 interaction events.

Streamlit Dashboard

Retention Cohort

User Segmentation

Session Funnel

Content Performance

📦 Datasets Used

Dataset	Source	Description
Last.fm User Listening History	Kaggle	User-artist play counts & timestamps
Spotify Podcast Metadata	Kaggle	Episode descriptions, ratings, categories
KKBox User Behavior	Kaggle	Session logs, subscription data
Synthetic Data (custom)	data/generate_synthetic_data.py	Simulated NOICE-style interaction logs

Demo scale: 2,000 users · 500 content items · 50,000 interaction events (synthetic)

All datasets used for educational and portfolio purposes only.

🔍 Analysis Chapters

📖 Chapter 1 — Content Consumption Patterns

"What are users actually listening to, and for how long?"

• Top genres by total playtime vs. by unique listener count
• Content length sweet spot — what duration drives highest completion rate?
• Format comparison — podcast episodes vs. music vs. live streams
• Discovery channels — how users find new content

💡 Key Finding: Content under 25 minutes achieves higher completion rates than long-form content — suggesting a strong case for short-form audio strategy.

📖 Chapter 2 — User Behavior & Segmentation

"Not all users are equal — who are your most valuable listeners?"

• RFM Segmentation to identify user tiers
• Power users vs. casual listeners and content preferences
• Time-of-day patterns by day and segment
• Device & platform breakdown for mobile vs. desktop behavior

💡 Key Finding: Power Listeners account for a disproportionate share of total platform playtime — retention of this segment is critical.

📖 Chapter 3 — Retention & Drop-Off Analysis

"When do users leave, and what are the warning signs?"

• Session funnel analysis for content journey drop-off
• Day-1, Day-7, Day-30 retention curves by cohort and acquisition channel
• Early churn indicators from first sessions
• Skip rate analysis by content type

💡 Key Finding: High first-session skip behavior is a strong early warning signal for 30-day churn risk.

📖 Chapter 4 — Content Strategy Opportunities

"What does the data say about where to invest next?"

• Underserved genres with high search demand and low content supply
• Creator performance analysis by repeat listener rate
• Cross-content journey between podcasts, music, and live streams
• Seasonal & trending patterns behind content spikes

💡 Key Finding: Talk Show and Tech content categories are underrepresented relative to search demand — a clear gap for content acquisition strategy.

📖 Chapter 5 — Executive Summary & Recommendations

"If you had to act on this data tomorrow, what would you do?"

Priority	Insight	Recommended Action	Expected Impact
🔴 High	High churn in first session	Improve onboarding recommendation flow	+15% Day-7 retention
🔴 High	Power users drive most playtime	Build loyalty program / exclusive content	Reduce power user churn
🟡 Medium	Short-form content outperforms	Commission more sub-25min content	+10% avg completion rate
🟡 Medium	Underserved Talk Show/Tech genres	Prioritize creator acquisition in these niches	Capture high-intent users
🟢 Low	Mobile peak hours 7–9PM	Schedule content drops and push notifications	+8% notification CTR

📊 Visualizations Included

Visual	Description
Genre Heatmap	Listening volume by genre × time of day
Retention Cohort Chart	Day-1/7/30 retention curves by cohort
RFM Segment Bubble Chart	User segments plotted by engagement value
Session Funnel	Drop-off at each stage of content consumption
Skip Rate by Category	Bar chart of skip rates across content types
Content Duration vs Completion	Scatter plot showing optimal content length
Creator Loyalty Index	Top creators ranked by repeat listener rate

🛠️ Tech Stack

Tool	Purpose
Pandas	Data wrangling & aggregation
Matplotlib / Seaborn	Static visualizations
Plotly	Interactive charts
Streamlit (optional)	Interactive dashboard version
Jupyter Notebook	Narrative + code storytelling format

⚙️ Setup & Installation

git clone https://github.com/LuthfiMirza/streaming-user-behavior-eda.git
cd streaming-eda-data-story
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

🚀 Running the Analysis

python data/generate_synthetic_data.py
jupyter notebook
# Or run the full story in one notebook
jupyter notebook notebooks/06_executive_summary.ipynb

💡 Why This Project Matters

Most ML portfolios jump straight to models. This project demonstrates something rarer and equally valuable: the ability to look at raw data, ask the right business questions, and communicate findings that non-technical stakeholders can act on.

For a streaming platform like NOICE, this translates directly to:

• Informing content acquisition strategy
• Improving recommendation engine inputs
• Guiding growth and retention campaigns
• Prioritizing product roadmap decisions

🛣️ Roadmap

• Synthetic dataset generation pipeline
• Content consumption analysis scaffolding
• User segmentation scaffolding
• Retention & drop-off analysis scaffolding
• Business recommendations framework
• Interactive Streamlit dashboard
• Bahasa Indonesia version of executive summary
• Integration with real public streaming datasets

👤 Author

Luthfi Mirza Darsono
Gunadarma University — Information Systems
📧 luthfimirza2004@gmail.com
🔗 LinkedIn | GitHub

📄 License

This project is licensed under the MIT License.