Market Basket & Sequential Pattern Analysis

Overview

This project analyzes customer purchasing behavior using descriptive data mining techniques. The objective is to uncover frequently co-purchased items and sequential buying patterns to support product bundling and recommendation strategies.

---

Dataset

• Online Retail II Dataset (UCI Repository)
• 1.04 million cleaned transactions
• Features: Invoice, StockCode, Description, Quantity, Price, Customer ID, Country

---

Objectives

• Identify frequent itemsets using Apriori
• Generate strong association rules
• Compare support & confidence thresholds
• Evaluate lift and rule strength
• Discover sequential purchasing patterns using PrefixSpan
• Analyze robustness across random seeds

---

Techniques Used

Association Rule Mining

• Apriori (multiple support/confidence experiments)
• FP-Growth
• Lift-based evaluation

Sequential Pattern Mining

• PrefixSpan algorithm
• Time-sorted customer transaction sequences

---

Key Findings

• Moderate thresholds (Support=2%, Confidence=50%) produced the best balance of rule quality and coverage.
• Average Lift ≈ 14 indicates strong co-purchase relationships.
• Sequential patterns show themed item purchasing behavior.
• Results are robust across multiple random seeds.

---

Business Impact

• Enables product bundling strategies
• Improves recommendation systems
• Supports inventory planning
• Identifies high-value item combinations

---

Technologies Used

• Python
• Pandas
• mlxtend
• PrefixSpan
• NumPy
• Matplotlib

---

How to Run

1. Install dependencies:

   pip install pandas mlxtend prefixspan
   

2. Open:

   purchase_pattern_analysis.ipynb
   

3. Run all cells sequentially.# customer-purchase-pattern-mining