XAI - Explainable Artificial Intelligence

This repository implements a movie-genre classification pipeline that extracts film metadata from DBpedia via SPARQL queries and uses multiple XAI techniques (SHAP, permutation importance, surrogate models) to interpret a Gradient Boosting classifier's decision-making process. The project demonstrates how director and actor metadata drive genre predictions between comedies and dramas, providing transparency into model behavior through complementary interpretability methods.

Setup Instructions

Prerequisites

• Python 3.12.0
• pip 24.2+ or pip 23.2.1
• Conda (recommended)

Installation

1. Clone the repository

   git clone https://github.com/Akshayaquinnox/XAI.git
   cd XAI

2. Create conda environment

   conda create -n xai-env python=3.12
   conda activate xai-env

3. Install required packages

   pip install -r requirements.txt

   Or install packages manually:

   pip install shap pandas numpy scikit-learn matplotlib jupyter rdflib sparqlwrapper

Running the Analysis

1. Start Jupyter Notebook

   jupyter notebook

2. Open the main notebook

   movie_explainability.ipynb
   

3. Run all cells to execute.

Project Structure

XAI/
├── movie_explainability.ipynb      # Main analysis notebook
├── requirements.txt                # Python dependencies
└── README.md                       # This file

Usage

The notebook demonstrates:

• DBpedia data extraction using SPARQL queries for movie metadata
• Feature engineering from RDF triples.
• Gradient Boosting classifier training for comedy vs drama classification
• Four interpretability techniques: Gini importance, permutation importance, SHAP values, and surrogate decision trees