E-Commerce Bag Image Classification System (Orange & Python)

This repository contains an image-based product classification project performed on a custom-collected dataset of bag images sourced from major e-commerce platforms (Trendyol, Amazon, eBay, Lindex, Shein).

• Author: Amra Kurtabegu
• Course: Data Mining

---

📊 Dataset Overview

The dataset consists of 4,000 balanced images organized into 4 distinct classes (1,000 images per class):

1. backpack (Backpack)
2. chest_bag (Chest / Sling Bag)
3. handbag (Handbag / Shoulder Bag)
4. laptop_bag (Laptop Briefcase Bag)

---

🛠️ Methodology & Workflow

1. Data Collection (Python): Custom scraper script (scraper.py) fetched raw product images with randomized delay intervals to handle server rate limits.
2. Data Preprocessing & Cleaning (Python & PIL): Filtered out corrupted files, removed MD5 duplicates, and balanced the dataset using preprocess.py.
3. Feature Extraction (Orange Image Analytics): Applied Image Embedding using pre-trained Deep Learning models to transform images into numerical feature vectors.
4. Model Evaluation: Evaluated machine learning algorithms using 10-Fold Cross Validation.

---

📈 Performance & Results

Model	Accuracy (CA)	Precision	Recall	F1-Score	AUC
kNN (k-Nearest Neighbors)	82.9%	83.0%	82.9%	82.8%	0.952
Logistic Regression	82.2%	82.2%	82.2%	82.2%	0.954
Naive Bayes	72.6%	73.8%	72.6%	72.8%	0.910

---

📁 Repository Structure

• scraper.py: Automated web scraper script for downloading raw product images.
• preprocess.py: Image validation, deduplication, and class balancing script.
• Veri Madenciligi Canta.ows: Orange Data Mining visual workflow pipeline file.
• README.md: Project documentation and performance summary report.

---

🚀 How to Run the Project

1. Install Orange Data Mining.
2. Install the Image Analytics add-on in Orange via Options -> Add-ons -> Image Analytics.
3. Load Veri Madenciligi Canta.ows in Orange.
4. Set the dataset path in the Import Images widget to point to your image directory.

---

📄 License

This project is licensed under the MIT License.