💍 Marriage Likelihood Prediction AI

Disclaimer: Fate and love are ultimately determined by God. This project is educational and explores how AI can analyze patterns of compatibility and relationship dynamics. It is not a verdict on human relationships and must be used responsibly.

Overview

This repository demonstrates the AI development workflow applied to a creative real-world problem: predicting the likelihood that a couple will marry based on quantified relationship features. It follows the CRISP-DM framework and includes ethical considerations, model evaluation, and reproducible notebooks.

Project structure

ai-marriage-likelihood/
├── data/
│   ├── raw/
│   └── processed/             # marriage_data.csv
├── models/                    # saved model artifact (marriage_model.pkl)
├── notebooks/
│   ├── 00-generate-data.ipynb
│   ├── 01-data-exploration.ipynb
│   ├── 02-model-training.ipynb
│   └── 03-evaluation.ipynb
├── src/
│   ├── data/preprocess.py
│   ├── models/train_model.py
│   ├── eval/evaluate.py
│   └── app/app.py
├── reports/
│   └── final_report.md
├── requirements.txt
├── Dockerfile
└── README.md

Installation (local)

git clone https://github.com/<your-username>/ai-marriage-likelihood.git
cd ai-marriage-likelihood
python -m venv venv
# Activate:
# Windows (Git Bash): source venv/Scripts/activate
# Windows (PowerShell): .
env\Scripts\Activate.ps1
# macOS / Linux: source venv/bin/activate
pip install -r requirements.txt

Usage (quick)

1. Generate synthetic data: open and run notebooks/00-generate-data.ipynb.
2. Explore data: run notebooks/01-data-exploration.ipynb.
3. Train model: run notebooks/02-model-training.ipynb or python src/models/train_model.py.
4. Evaluate: run notebooks/03-evaluation.ipynb or python src/eval/evaluate.py.
5. Sample prediction: python src/app/app.py (loads models/marriage_model.pkl).

Evaluation snapshot

Example classification results from a final test run (see notebooks/03-evaluation):

accuracy: 0.52
precision (low): 0.50, recall (low): 0.90
precision (high): 0.62, recall (high): 0.16

Interpretation: model is good at detecting couples unlikely to marry but weak at identifying true positives; further data balancing and richer features are recommended.

Ethics & Responsible Use

• Use only anonymized, consensual data.
• Clearly explain limitations and uncertainties to users.
• Avoid making life-changing decisions solely based on model outputs.
• Include opt-out and human review mechanisms in any future product.

Author

Bikila Keneni — AI for Software Engineering (2025)

---