🛒 Smart Sales Channel Prediction: Multi-Class Classification with XGBoost

🎯 Business Overview

In the modern retail ecosystem, understanding whether customers prefer Online, Offline, or Hybrid shopping channels is essential for improving marketing efficiency and customer engagement. This project builds a multi-class classification model to predict customer shopping channel preferences using digital behavior data and demographic characteristics.

With this model, retail companies can:

• Optimize marketing strategies based on customer preferences
• Allocate promotions more effectively between online and offline channels
• Better understand customer behavior in an omnichannel retail ecosystem

📊 Dataset

Dataset source:
https://www.kaggle.com/datasets/shree0910/online-vs-in-store-shopping-behaviour-dataset

The target variable in this project is: shopping_preference

• Store → customers primarily shop in physical stores
• Online → customers primarily shop through online platforms
• Hybrid → customers use both online and offline shopping channels

The dataset contains 11,789 customers with multiple features describing shopping behavior, digital activity, and product interaction preferences.

Label Distribution

The label distribution shows that the Store class dominates the dataset, compared to Online and Hybrid classes. This imbalance presents a challenge during modeling because the algorithm may become biased toward the majority class.

🤖 Machine Learning Model

The model used in this project is XGBoost Classifier, a Gradient Boosting-based algorithm well known for its strong performance in classification tasks.

Modeling Strategy

• Primary Algorithm: XGBoost Classifier
• Imbalanced Data Handling: compute_class_weight
• Hyperparameter Optimization: RandomizedSearchCV
• Main Evaluation Metric: F1-Score

This approach ensures that the model not only achieves high accuracy but also performs well across imbalanced classes.

📊 Model Evaluation

Model performance was evaluated using a classification report and confusion matrix.

Model Accuracy 96%

Classification Report

Class	Precision	Recall	F1-score
Store	0.99	0.98	0.99
Online	0.91	0.93	0.92
Hybrid	0.41	0.51	0.46

Model Performance Analysis

• The model performs extremely well in predicting Store and Online classes.
• Performance on the Hybrid class is lower, likely because Hybrid customers exhibit mixed behavior between online and in-store shopping.
• The Weighted F1-score of 0.96 indicates strong overall classification performance.

🧠 Analysis & Insights

Based on the modeling results and feature interpretation using SHAP (SHapley Additive exPlanations), several key insights were identified.

1. Digital Maturity

Features such as tech_savvy_score and daily_internet_hours are among the strongest predictors of customer shopping preferences.

Customers with low technological familiarity are rarely predicted as Online shoppers.

2. Hybrid Behavior Pattern

Customers in the Hybrid category are highly sensitive to shipping_cost_sensitivity.

• When shipping costs increase → customers tend to switch to Offline shopping
• When discounts are offered → customers remain Online shoppers

This suggests that Hybrid customers are highly responsive to promotional strategies.

3. Income Influence

The feature monthly_income indicates that customers with higher income levels are more likely to become Hybrid shoppers.

This reflects a flexible lifestyle that balances the convenience of online shopping with the experience of visiting physical stores.

4. Probabilistic Threshold

Probability distribution analysis shows that the Hybrid class has a thinner classification boundary compared to the other classes.

As a result, Hybrid customers require more targeted marketing strategies to convert them into Online-first customers.