Product Recommender System

📝 Introduction

This project aims to develop an association rule-based recommender system, using customer purchase transactions from a supermarket. The goal is to identify consumption patterns that allow us to recommend products frequently bought together.

We use the Apriori and FP-Growth algorithms to find frequent itemsets and generate association rules, filtered based on metrics such as support, confidence, and lift.

---

1. Clone the repository and install dependencies:

   git clone https://github.com/higorfct/Recommender-System
   cd Recommender-System
   pip install -r requirements.txt

---

📊 Data

The data is contained in the transactions.csv file, which consists of lists of products purchased by customers in a single transaction. Each line represents an individual transaction.

Steps performed:

• Reading and cleaning transactions (removal of spaces and quotes)
• Transformation to transactional format (one-hot encoding with TransactionEncoder)
• Generation of frequent itemsets with Apriori and FP-Growth
• Extraction of rules with association_rules()
• Application of filters based on support, confidence, and lift

---

🤖 Modeling with Apriori and FP-Growth

Modeling was done using two popular algorithms for association pattern analysis:

• Apriori
• FP-Growth

The main parameters used were:

• Minimum support: 1%
• Minimum confidence: 20%
• Minimum lift: 1.5
• Minimum size of the antecedent (LHS): 1

The rules generated were ordered by lift to highlight the most relevant ones.

---

🔍 Examples of Generated Rules

Antecedent	Consequent	Confidence	Lift
[milk]	[coffee]	0.45	1.8
[bread, butter]	[milk]	0.37	1.7
[soda]	[chips]	0.29	1.6

These rules indicate products that frequently appear together in transactions and can be used for targeted recommendations.

---

💼 Estimated Financial Impact

This script estimates the potential financial impact of a recommender system, considering parameters defined by the business.

Parameters

• Average profit per transaction: R$ 5.00 (≈ $1.00)
• Estimated increase in conversion rate: 10%
• Total transactions: 500

Calculations

• Additional transactions = 500 × 0.10 = 50
• Financial impact = 50 × R$ 5.00 = R$ 250.00 (≈ $50.00)

Results

• Estimated additional transactions: 50
• Estimated financial impact: R$ 250.00 (≈ $50.00)

This gain can be even greater with continuous personalization and real-time integration of recommendations.

---

📈 Visualizations

• Table with frequent itemsets found
• Sorted table with association rules filtered by lift
• (Optional) Bar charts of the most frequent items or common pairs

---

🛠️ Tools Used

• Python – Main programming language
• Pandas – Data manipulation
• MLxtend – Apriori, FP-Growth algorithms and rule generation
• Google Colab – Development environment

---

✅ Results

• Various association rules between products were generated.
• The system can provide automatic recommendations based on purchase history.
• The strongest rules have high lift and can serve as a basis for promotional actions or suggestions in e-commerce.

---

🧠 Conclusions

The project demonstrates how frequent pattern mining techniques can:

• Identify products with strong purchase correlation
• Support cross-selling strategies
• Improve personalization in recommendation platforms

---

🔄 Next Steps

• Implement real-time recommendations based on the user's cart
• Integrate with customer profile data for personalized recommendations
• Test rules in an A/B Testing system to measure impact on sales

---

🧑‍💻 Author and Contact

Higor Roberto Coutinho Caetano
LinkedIn: https://www.linkedin.com/in/higor-caetano-049521136/
E-mail: higorfct@gmail.com