Khush Libaas – AI-Powered Fashion Recommendation System

This repository contains all FAIR-related artifacts for the Khush Libaas experiment, created as part of the 2025 FAIR Data Science Exercise at TU Wien.

Project Overview

Khush Libaas is a personalized fashion recommendation system for Pakistani users. It leverages structured metadata (brand, color, fabric, style, etc.) to classify traditional outfits by event type (e.g., Wedding, Eid, Formal).

For Part-2, a simulated KNN-based pipeline was created, with all data and outputs made FAIR: Findable, Accessible, Interoperable, and Reusable.

Dataset PIDs on DBRepo (TU Wien)

| Dataset Type | PID | |--------------|-----| | Training | https://test.dbrepo.tuwien.ac.at/pid/9b858643-5ff0-495a-bcef-8c2e77d85713 | | Validation | https://test.dbrepo.tuwien.ac.at/pid/0a1a4634-38cc-486a-9e4c-746819d350b4 | | Test | https://test.dbrepo.tuwien.ac.at/pid/2e5a8c3f-9072-4d6d-ae5d-4d0342127cba |

Each dataset contains metadata including brand, price, size, color, gender, fabric type, and regional popularity.

Outputs Included

| Filename | Description | |----------|-------------| | khush_libaas_model.pkl | Trained KNN model file (simulated on synthetic data) | | evaluation_metrics.json | Evaluation metrics (accuracy, precision, recall, F1) | | confusion_matrix.png | Visualization of true vs predicted event types | | feature_importance_chart.png | Frequency-based importance of encoded features | | recommendations.csv | Top-5 recommended clothing samples with scores |

Tools & Technologies

  • Language: Python
  • ML Model: K-Nearest Neighbors (KNN)
  • Libraries: scikit-learn, pandas, matplotlib, seaborn
  • Notebook: Jupyter (notebook included in GitHub submission)

FAIR Compliance

This project complies with FAIR principles via:

  • Data deposition on DBRepo (TU Wien)
  • Output metadata uploaded to TUW Research Data
  • Metadata models used: FAIR4ML, CodeMeta, Croissant
  • All datasets have valid PIDs and are cited in the notebook

Licensing

  • Code & Scripts: MIT License
  • Data & Visuals: Creative Commons Attribution 4.0 International (CC BY 4.0)

Author

Mohsin Khalid
TU Wien — Matriculation No: 12443164

Contact

For questions, contact via TU Wien email.