🛒 Amazon Recommendation System: Applied AI Project

---

📋 Project Overview

This project was developed as part of the MIT Professional Education – Applied AI & Data Science Program, specifically within the Elective Project: Recommendation Systems.

The objective is to design, implement, and evaluate multiple recommendation system approaches using an Amazon product ratings dataset.

---

🎯 Objective

To build a recommendation system capable of suggesting relevant products while addressing:

• Data sparsity
• Popularity bias
• Cold-start problem

---

📊 Dataset

Feature	Description
user_id	Unique user identifier
prod_id	Unique product identifier
rating	Rating (1–5)
timestamp	Not used

🔗 Dataset:
👉 https://drive.google.com/file/d/1XahZcR287ke7j48I7-oj0KzmmwSSvA3Y/view

---

🧹 Data Preprocessing

• Users ≥ 50 ratings
• Products ≥ 5 ratings

➡️ Reduced sparsity and improved reliability

---

🔍 Exploratory Data Analysis

⭐ Rating Distribution

👤 User Activity

📦 Item Popularity

📉 Long Tail Distribution

🔝 Top Users

---

🤖 Models Implemented

📌 Rank-Based Model

• Non-personalized baseline

👥 User-User Collaborative Filtering

• Cosine similarity
• Personalized

📦 Item-Item Collaborative Filtering

• More stable than user-user

🧮 Matrix Factorization (SVD)

• Handles sparsity effectively
• Best performance

---

📈 Model Comparison

Model	RMSE	Precision	Recall	F1
UU Baseline	1.001	0.855	0.867	0.861
UU Optimized	0.953	0.847	0.894	0.870
II Baseline	0.995	0.851	0.870	0.860
II Optimized	0.958	0.839	0.880	0.859
SVD Baseline	0.888	0.853	0.880	0.866
SVD Optimized	0.882	0.860	0.900	0.879

---

📊 Model Performance Visualization

🔥 Heatmap

📊 Precision & Recall

📉 RMSE Comparison

---

📊 Prediction Analysis

📉 Error Distribution (All Models)

📍 SVD Predictions

📍 User-User Predictions

📊 Error (User-User)

---

📍 Recommendation Behavior

---

⚠️ Key Challenges

• High sparsity (~99%)
• Popularity bias
• Cold-start problem

---

🏆 Key Results

• Rank-based model provides a strong baseline
• KNN improves personalization but struggles with sparsity
• Item-item is more stable
• SVD achieves the best overall performance

---

📌 Conclusion

Matrix factorization (SVD) provides the best balance between:

• Accuracy
• Robustness
• Scalability

---

🛠️ Technologies

• Python
• Pandas / NumPy
• Matplotlib / Seaborn
• Scikit-surprise

---

🌟 Highlights

• End-to-end recommendation system pipeline
• Advanced EDA
• Model comparison
• Hyperparameter tuning
• Visual evaluation

---

👨‍💻 Author

Developed as part of the
MIT Professional Education – Applied AI & Data Science Program

---

📄 License

Educational use only