🏳️ Religious Classification Using Exploratory Data Analysis and Neural Networks

Predicting the dominant religion of a country using flag design features and geopolitical attributes with EDA and deep learning.

---

📚 Overview

This project explores the relationship between a country's flag design and its dominant religion, leveraging exploratory data analysis (EDA) and neural network models. Using both visual (image-based) and numerical (CSV-based) data, we implement and evaluate models like custom CNNs, MobileNetV2, and multi-layer perceptrons to classify religion across countries.

---

🧠 Introduction

🎯 Research Focus

The core idea is to evaluate whether visual elements of national flags can be used to predict a country's dominant religion.

🧾 Objectives

• Conduct EDA on geopolitical and flag-related features.
• Build ML models that classify religion based on:
  • Flag images (custom CNN, MobileNetV2)
  • Geopolitical numeric data (Multi-layer Neural Network)
• Compare model performance and draw conclusions.

---

📊 Dataset

The dataset includes 177 countries with 30+ features, such as:

• Geopolitical attributes: continent, population, area, language, religion
• Flag design features: number of colors, dominant color, presence of symbols (crosses, stars, crescents, suns, circles)

🖼️ Flag Images

A custom dataset of flag images (from 1986) is compiled and aligned with the CSV metadata.

---

🧪 Methodology

1. Data Cleaning

• Removed irrelevant files and images
• Mapped image names to countries
• Standardized categories and labels

2. Exploratory Data Analysis (EDA)

• Distribution of colors, symbols, and religion types
• Correlation matrices (e.g., between symbols and religion)
• Comparative visualizations across continents and religions

3. Model Development

Model Type	Input	Description
Custom CNN	Flag Images	Designed from scratch to classify religion
MobileNetV2	Flag Images	Pre-trained model adapted to our dataset
MLP	CSV Data	Neural network trained on numerical features

---

📈 Results

🔍 EDA Insights

• Red and white are the most frequent flag colors globally.
• Christian countries often feature crosses on their flags.
• Flags of Muslim-majority countries frequently feature green and crescent symbols.

🤖 Model Performance

Model	Accuracy (Global)	Accuracy (Africa)
Custom CNN	50%	80%
MobileNetV2	20%	20%
MLP (CSV)	56% (peak)	-

⚠️ Note: Low accuracy is due to limited dataset size and visual diversity among flags.

---

📌 Conclusion & Future Work

🔎 Key Takeaways

• Flag design does show some correlation with religion, but it's not conclusive.
• Dividing countries by region improves model accuracy.
• Complex geopolitical and historical factors limit prediction reliability.

🚀 Future Improvements

• Use higher-resolution and modern flag images
• Expand dataset for better generalization
• Add contextual geopolitical data
• Fine-tune and ensemble models

---

🛠️ Setup & Usage

📋 Prerequisites

• Python 3.8+
• Libraries:
  tensorflow, keras, pandas, numpy, matplotlib, seaborn, scikit-learn

---

▶️ Running the Project

You have two options for running the project:

✅ Option 1: Run on Google Colab

• Simply upload the main code file and the datasets/ folder (containing flag_images/ and flags_csv/) to your Colab session.
• Then, run the cells in the provided notebooks for EDA and data preprocessing.
• You can also train models (CNN, MobileNetV2, MLP) by running the corresponding code cells in the notebook.

💻 Option 2: Run Locally

• Clone the GitHub repository:

  git clone https://github.com/your-username/religious-classification.git
  cd religious-classification

• Make sure your environment is set up (e.g., with requirements.txt or using conda/venv).
• Run the notebook for EDA and preprocessing.
• For model training, run the script cells in the provided notebooks.

---

📂 Dataset Preparation

• Place your datasets in the following structure:

religious-classification/
├── datasets/
│   ├── flag_images/
│   └── flags_csv/

📚 References

• UCI Flags Dataset
• Flag Images: Flaglog 1986
• TensorFlow & Keras Documentation
• MobileNetV2 Paper