Predicting Customer Spending Patterns on E-commerce Websites using SVM's

Project Overview

This project uses a Support Vector Machine (SVM) with a polynomial kernel to predict if a website visit led to a purchase. It analyzes session data and uses GridSearchCV to fine-tune the model for better accuracy.

Key Features

• SVM Classifier with Polynomial Kernel: Used for classification of visitor sessions.
• Hyperparameter Tuning with GridSearchCV: Optimizes model parameters for better accuracy.
• Data Preprocessing Pipelines: Handles numerical and categorical feature transformations.
• Permutation Importance Analysis: Assesses feature impact on predictions.
• Model Evaluation & Visualization: Includes confusion matrix, learning curves, and feature distributions.

Files in the Project

• ecommerce_notebook.ipynb: Jupyter notebook containing the complete implementation.
• requirements.txt: List of dependencies required to run the project.

How to Run

1. Ensure you have Python installed.
2. Install the required dependencies:

   pip install -r requirements.txt

3. Open and run ecommerce_notebook.ipynb in Jupyter Notebook or execute the script in a Python environment.

Steps in the Project

1. Import Necessary Libraries
2. Load and Split the Data: Uses train_test_split for splitting.
3. Preprocess Data: Uses pipelines for numerical and categorical features.
4. Define Model and Hyperparameter Grid: Sets up SVM classifier with polynomial kernel.
5. Train Model with GridSearchCV: Optimizes parameters through cross-validation.
6. Evaluate Model on Test Set: Computes accuracy and feature importance.
7. Visualizations:
   • Feature distributions (numerical & categorical)
   • Correlation heatmap
   • Confusion matrix
   • Learning curve for training sample sizes

Results

• Best hyperparameters selected using GridSearchCV.
• Training and test accuracy scores.
• Feature importance ranking via permutation importance.
• Learning curve visualization showing model performance over different sample sizes.

Conclusion

This project provides a structured approach to classifying website visitor sessions using an SVM classifier with a polynomial kernel. Through hyperparameter tuning and feature importance analysis, it offers insights into the most relevant factors influencing purchase decisions.

Author

Matthew Neba