IMDb Score Prediction Project

Overview

This project classifies films into four IMDb score categories (Bad, OK, Good, Excellent) using movie metadata, financial data, and social media metrics. We employ Decision Trees, K-Nearest Neighbors (KNN), and Random Forests to predict critical reception and understand key drivers of success.

---

Dataset

Source: Kaggle IMDb 5000 Movie Dataset

Location: movie_metadata.csv

Key Features:

• Target: imdb_score (binned into 4 categories)
• Financial: budget, gross
• Metadata: duration, year, country, content_rating
• Social Metrics: movie_facebook_likes, director_facebook_likes, actor_1_facebook_likes, actor_2_facebook_likes, actor_3_facebook_likes
• Derived: critic_review_ratio, other_actors_facebook_likes

---

Installation

1. Clone the repository

   git clone https://github.com/tanayprabhakar/FDSProj.git
   cd FDSProj

2. Install R dependencies

   install.packages(c(
     "dplyr",
     "ggplot2",
     "caret",
     "randomForest",
     "rpart",
     "ggrepel",
     "VIM",
     "plotly"
   ))

3. Project structure

   FDSProj/
   │── movie_metadata.csv
   │── data_preprocessing.R
   │── rf_model.rds
   │── knn_model.rds
   ├── README.md
   

---

Usage

1. Run the full analysis

   # in R console
   source("data_preprocessing.R")

---

Workflow

1. Data Preprocessing

• Removed duplicates and handled missing values
• Binned imdb_score into four categories: Bad (<=5), OK (5–6.5), Good (6.5–8), Excellent (>8)
• Engineered critic_review_ratio (num_critic_for_reviews / num_user_for_reviews)
• Combined actor_2_facebook_likes and actor_3_facebook_likes into other_actors_facebook_likes
• Simplified country to USA/UK/Others
• Standardized content_rating levels

2. Exploratory Data Analysis (EDA)

• High-budget (> $100M) films seldom achieve "Excellent" ratings
• UK films average 0.3 points higher IMDb scores than US films
• Weak correlation between Facebook likes and IMDb scores (r = 0.18)

3. Modeling

Model	Training Accuracy	Validation Accuracy	Test Accuracy
Decision Tree	78.0%	71.3%	72.4%
KNN (k = 9)	—	71.4%	74.6%
Random Forest	—	76.4%	76.6%

Top Features (Random Forest):

1. num_voted_users (Importance: 42.6)
2. duration (38.1)
3. critic_review_ratio (35.8)

---

Results Interpretation

1. Critic Influence: Films with critic_review_ratio > 0.5 are 3× more likely to be Excellent.
2. Ideal Runtime: 100–120 minutes yields the highest average score (7.2).
3. Country Effect: UK productions score on average 0.4 points higher than US films.

---

Contributing

Contributions are welcome! Please fork the repository and create a pull request for bug fixes or feature requests.

---

Contact

Maintainer: Tanay & Tarang Last Updated: May 2025