SmartCart-Recommendation-Analysis

A recommendation system analysis for e-commerce platforms using collaborative filtering and association rule mining.

Overview

This project implements and evaluates two main recommendation approaches:

• User-based collaborative filtering for personalized product recommendations
• Association rule mining to discover purchasing patterns

The analysis uses real e-commerce user interaction data to build and compare different recommendation strategies.

Contributors

Name	Student ID	GitHub
Antoine Cantin	40211205	@ChiefsBestPal
Tuan Anh Pham	40213926	@TuanAnh-P
Ryan Li	40214839	@Ryan2Li
Mustafa Sameem	40190889	@MustafaSameem

Features

Collaborative Filtering

• Cosine similarity between users
• Top-N product recommendations
• Performance evaluation with Precision@K and Recall@K
• Grid search for parameter optimization

Association Rule Mining

• Apriori algorithm implementation
• Frequent itemset discovery
• Rule evaluation using support, confidence, and lift

Data Analysis

• Exploratory data analysis
• User behavior and product popularity analysis
• Visualization of patterns and results

Getting Started

Requirements

• Python 3.8+
• See requirements.txt for package dependencies

Installation

1. Clone this repository
2. Install dependencies: pip install -r requirements.txt
3. Open the Jupyter notebook: jupyter notebook analysis.ipynb

Usage

Run the notebook cells in order to:

1. Load and explore the dataset
2. Build user-item matrices
3. Train collaborative filtering models
4. Generate recommendations
5. Mine association rules
6. Evaluate and visualize results

Project Structure

SmartCart-Recommendation-Analysis/
├── data/
│   ├── ecommerce_user_data.csv    # User ratings and interactions
│   └── product_details.csv        # Product metadata
├── analysis.ipynb                 # Main analysis notebook
├── report.pdf                     # Analysis report
├── README.md
└── requirements.txt

Data

The dataset contains:

• User interaction data with ratings, product IDs, and categories
• Product metadata including names and categories
• Approximately 700+ user interactions across 100+ products

Methods

Data Preprocessing

• User-item matrix construction
• Missing value handling
• Train/test splitting (80/20)

Collaborative Filtering

• Mean-centered user-item matrix
• Cosine similarity calculation
• K-nearest neighbors approach
• Weighted rating prediction

Association Rules

• Transaction format conversion
• Frequent itemset mining with configurable support thresholds
• Rule generation with confidence and lift metrics

Results

The analysis achieves:

• Precision@10 around 0.65 for collaborative filtering
• Discovery of meaningful product associations
• Coverage of 80%+ of catalog items in recommendations
• Matrix sparsity typical of e-commerce datasets (~95%)

Configuration

Key parameters can be adjusted:

K_RECOMMENDATIONS = 10      # Recommendations per user
TOP_K_USERS = 10           # Similar users to consider
MIN_SUPPORT = 0.05         # Minimum support for itemsets
MIN_CONFIDENCE = 0.5       # Minimum confidence for rules

Dependencies

• pandas: Data manipulation
• numpy: Numerical operations
• scikit-learn: Machine learning tools
• matplotlib/seaborn: Visualization
• mlxtend: Association rule mining
• jupyter: Notebook environment

License

MIT License