🎬 Netflix Content Analysis — EDA, Statistics & Probability

An end-to-end data analysis project on Netflix's content library using Python — covering data cleaning, exploratory data analysis, statistical hypothesis testing, probability distributions, and advanced analytical deep dives.

---

📌 Project Overview

This project performs a comprehensive analysis of Netflix's content catalog (netflix_titles.csv) to uncover patterns in content type, genre distribution, country contributions, release trends, and viewing ratings. The analysis progressively builds from basic data cleaning to advanced statistical reasoning and probabilistic modelling.

Dataset: Netflix Movies and TV Shows — Kaggle
Records: ~8,800 titles | Features: 12 columns
Tools: Python, Pandas, NumPy, Matplotlib, Seaborn, SciPy, WordCloud

---

🗂️ Project Structure

netflix-content-analysis/
│
├── data/
│   ├── netflix_titles.csv        ← Raw dataset (download from Kaggle)
│   └── netflix_cleaned.csv       ← Cleaned dataset (output after preprocessing)
│
├── images/                       ← All plot outputs saved here
│   └──....
├── notebook/
│   └── Netflix_data.ipynb        ← Main analysis notebook
│
└── README.md

---

📊 Analysis Sections

1. Data Cleaning

• Handled missing values in director, cast, country, date_added
• Parsed date_added to extract year_added and month_added
• Engineered features: duration_num, duration_unit, genre_list
• Removed duplicates on (title, release_year)
• Standardised text columns (type, rating)

2. Exploratory Data Analysis (EDA)

Analysis	Key Finding
Movies vs TV Shows	~70% of content is Movies
Top countries	USA dominates, followed by India & UK
Rating distribution	TV-MA is the most common rating
Top genres	International Movies, Dramas, Comedies lead
Top directors & actors	Rajiv Chilaka, Anupam Kher among top contributors
Movie duration	Right-skewed distribution, mean ~99 min
Monthly trends	January & July see peak content additions

3. Probability Measures

Applied real-world probability distributions to Netflix data:

• Bernoulli — P(randomly picked title is a Movie) = 0.696
• Binomial — P(exactly 7 out of 10 random picks are Movies)
• Poisson — Modelling average new content arrivals per year (λ ≈ 119)

4. Statistical Hypothesis Testing (Tier 1)

Test	Question	Result
Independent t-test	Is movie duration significantly different from TV show seasons?	Reject H₀ (p < 0.05)
Shapiro-Wilk + QQ-Plot	Are movie durations normally distributed?	Not normal — right skewed
Chi-Square test	Is there an association between content type and rating?	Reject H₀ — significant association
WordCloud	Most frequent words in Netflix descriptions	family, love, life, friend…

5. Advanced Statistical Analysis (Tier 2)

• Conditional Probability — P(genre | country): Each country's content personality mapped as a probability heatmap
• Central Limit Theorem Demo — Showed how sample means of skewed movie durations converge to normal at n ≥ 30
• Outlier Detection — IQR, Z-Score, and Modified Z-Score methods compared on movie durations
• Content Gap Analysis — Identified underserved genres per country vs. global average
• YoY Growth Rate + CAGR — Computed Compound Annual Growth Rate; peak absolute additions in 2019
• Multi-Variable EDA — Pair plots, correlation heatmap, hexbin density (duration vs release year), content lag distribution

6. Growth Rate Deep Dive

Investigated why Netflix's YoY growth rate declined from 423% (2016) to -20% (2021):

• Base Effect — % growth naturally falls as cumulative base grows
• Absolute additions — Peaked in 2018–2019, genuine slowdown after
• COVID Impact — Monthly analysis revealed production halt in Mar–Jun 2020
• Strategy shift — Netflix moved from quantity licensing to quality originals post-2020

---

📈 Key Insights

1. Netflix is movie-heavy — 69.6% of titles are Movies, but TV Show additions grew faster post-2018
2. USA + India drive the catalog — Together accounting for over 40% of all content
3. Content lag is short — Median gap between a movie's release and its Netflix addition is just 1–2 years
4. Rating skews mature — TV-MA and R dominate; Kids/Family content is proportionally underrepresented
5. Growth % is misleading alone — 423% in 2016 is a base-effect illusion; absolute additions peaked in 2019

---

🛠️ Tech Stack

Python 3.x
├── pandas          — data manipulation
├── numpy           — numerical computing
├── matplotlib      — base visualisations
├── seaborn         — statistical plots
├── scipy.stats     — hypothesis testing, distributions
└── wordcloud       — text frequency visualisation

---

⚙️ Setup & Run

# 1. Clone the repo
git clone https://github.com/byteephantom/netflix-content-analysis.git
cd netflix-content-analysis

# 2. Install dependencies
pip install pandas numpy matplotlib seaborn scipy wordcloud

# 3. Download dataset
# Place netflix_titles.csv inside the data/ folder
# Dataset: https://www.kaggle.com/datasets/shivamb/netflix-shows

# 4. Launch notebook
jupyter notebook notebook/Netflix_data.ipynb

---

🔭 Roadmap — What's Coming Next

• K-Means Clustering — Group Netflix content into natural content profiles
• Classification Model — Predict Movie vs TV Show from features (Logistic Regression + Random Forest)
• TF-IDF Content Recommender — "You watched X → here are similar titles"
• Sentiment Analysis — VADER on descriptions; do thrillers use more negative language?
• Streamlit App — Interactive dashboard with live recommender, deployed on Streamlit Cloud

---

📁 Dataset

Source: Kaggle — Netflix Movies and TV Shows
The dataset is not included in this repository. Download netflix_titles.csv from Kaggle and place it in the project root before running the notebook.

---

👤 Author

Quantum Nomad (@byteephantom)

📄 License

This project is open source and available under the MIT License.