🛒 Customer Segmentation

📌 Business Context

Instacart, a U.S.-based online grocery delivery platform, processes millions of orders across thousands of retailers. A core business challenge is increasing user retention and order volume through personalized recommendations and targeted marketing. With more than 3 million anonymized grocery orders from 200,000+ users, this project explores how data science can drive these outcomes by understanding customer behavior and predicting future purchases.

🎯 Project Objectives

• 🔍 Understand purchase behavior through exploratory data analysis (EDA).
• 🧮 Segment customers for targeted marketing using clustering techniques.
• 🤖 Predict product reorders using supervised ML models on user and product-level features.
• 🧠 Interpret model decisions using explainability tools like SHAP.

🗂️ Project Structure

├── Clustering/                                  
│ ├── Data Description and EDA.ipynb          # Exploratory Data Analysis (EDA)  
│ ├── Customer Segmentation.ipynb             # Customer segmentation based on product aisles  
├── Prediction/                           
│ ├── Data Preparation.ipynb                    # Dataset merging and transformation
│ ├── Feature Extraction.ipynb                  # Engineering 30+ features from user/product behavior
│ ├── ANN Model.ipynb                           # Neural network for reorder prediction
│ ├── XGBoost Model.ipynb                       # XGBoost with SHAP-based model interpretation

🧠 Methodology

1. Exploratory Data Analysis

• Combined 6 raw datasets (~3M orders) into a unified memory-efficient structure.
• Identified top aisles, reorder rates, cart sizes, and weekly order patterns.

Key Insights:

• 60%+ of products are reorders.
• Organic and fresh produce are top repeat buys.
• Most orders occur on weekends, especially Saturday afternoons.

2. Customer Segmentation

• Created a customer–aisle matrix and reduced dimensionality using PCA.
• Applied KMeans clustering to uncover 5 distinct customer personas.
• Enabled targeted campaigns for each segment.

3. Feature Engineering

Extracted 30+ features across:

• Product-level: reorder rates, popularity, organic flag
• User-level: order frequency, diversity, reorder percentage
• User-product interaction: historical behavior
• Aisle/department: volume, recency, encoded categories

4. Predictive Modeling

• Trained XGBoost and ANN models to predict product reorder likelihood.
• Applied class weighting to handle class imbalance.
• Evaluated models using ROC-AUC and SHAP for explainability.

📊 Results & Insights

Model	ROC-AUC	Key Insights
XGBoost	~0.86	Top features: User reorder ratio, product popularity
Neural Net	~0.85	Comparable performance with nonlinear feature handling

✅ SHAP values highlighted important behavioral signals and made predictions explainable.

🚀 Business Impact

• 📦 Improved reorder prediction enables smarter cart suggestions.
• 🛍️ Association rules support bundle recommendations and shelf optimization.
• 👥 Customer segmentation powers personalized outreach and lifecycle marketing.

📥 How to Use

1. Clone the repository

git clone https://github.com/yourusername/instacart-order-prediction.git
cd instacart-order-prediction

2. Download the dataset from Kaggle and place the CSV files in the data/ directory.

  Run notebooks in order:

- Clustering/Data Description and EDA.ipynb

- Clustering/Customers Segmentation.ipynb

- Prediction/Data Preparation.ipynb

- Prediction/Feature Extraction.ipynb

- Prediction/XGBoost Model.ipynb or ANN Model.ipynb