Netflix Content Analysis & Recommendation System

End-to-end analysis of the Netflix catalogue — EDA, genre analysis, a collaborative-filtering recommender, and IMDb rating prediction.

---

Dashboard Overview

---

Repository Structure

netflix-analysis/
├── netflix_analysis.py   # Python implementation (self-contained, demo data)
├── netflix_analysis.R    # Cleaned R script (requires real datasets)
├── Outputs/              # All generated plots
└── README.md

---

Quick Start

Python (no real data needed — runs on synthetic demo data)

pip install pandas numpy matplotlib seaborn scikit-learn wordcloud scipy
python netflix_analysis.py

R (requires the three source CSV files)

# Datasets needed in working directory:
#   netflix_titles.csv  ·  IMDb ratings.csv  ·  IMDb movies.csv

source("netflix_analysis.R")

---

Analysis Sections

#	Section	Key Output
1	Data Loading & Merging	Netflix + IMDb joint dataset
2	Preprocessing	Deduplication, mode-filling, date parsing
3	EDA & Visualisations	9 plots (see below)
4	Recommendation System	IBCF top-10 per user
5	Predictive Modelling	Linear Regression + Gradient Boosting
6	TF-IDF Text Analysis	Top terms per title

---

Visualisations

1 · Content Type Split

70 % Movies vs 30 % TV Shows on the platform.

---

2 · Top 10 Countries

United States dominates, followed by India and the United Kingdom.

---

3 · Content Growth Over Time

Rapid library expansion between 2015 and 2020.

---

4 · Rating Distribution

TV-MA and TV-14 account for the majority of content.

---

5 · Top 20 Genres

Drama and Comedy are the most frequent categories.

---

6 · Movie Duration by Country

Indian movies tend to be the longest; South Korean titles the shortest.

---

7 · Genre Co-occurrence Heatmap

Drama + Comedy and Drama + Thriller are the most common genre pairings.

---

8 · Top Directors & Actors

---

9 · Word Frequency in Descriptions

---

Recommendation System

Item-Based Collaborative Filtering (IBCF) using cosine similarity on a user × item rating matrix. The model finds the k most similar items to those a user has already rated and surfaces the highest-scoring unseen titles.

Item Similarity Heatmap (first 20 items)

Sample recommendations for User 0:

 1. Title_257    6. Title_135
 2. Title_98     7. Title_179
 3. Title_107    8. Title_258
 4. Title_96     9. Title_10
 5. Title_176   10. Title_285

---

Predictive Modelling

Two models are trained to predict IMDb weighted_average_vote from metadata features:

Model	RMSE
Linear Regression	1.23
Gradient Boosting	1.25

Feature Importance (Gradient Boosting)

Actual vs Predicted Ratings

---

TF-IDF Analysis

TF-IDF scores surface the most distinctive words per title, beyond simple frequency.

---

Tech Stack

Component	Python	R
Data wrangling	pandas, numpy	tidyverse, data.table
Visualisation	matplotlib, seaborn	ggplot2, plotly
Recommendation	scikit-learn (cosine sim)	recommenderlab (IBCF)
ML modelling	scikit-learn (LR + GBM)	caret, gbm
Text analysis	scikit-learn TF-IDF	tm, tidytext

---

Data Sources

Dataset	Source
Netflix Titles	Kaggle — Netflix Movies and TV Shows
IMDb Ratings	Kaggle — IMDb movies extensive dataset

Note: The Python script ships with synthetic demo data so it runs without any downloads.