Skip to content

Latest commit

 

History

History
34 lines (29 loc) · 2.29 KB

README.md

File metadata and controls

34 lines (29 loc) · 2.29 KB
Raw

INSY 695 Final Project

Group 3: Fourchetteurs

Data source: https://www.kaggle.com/datasets/berkayalan/bank-marketing-data-set

Team Members (Enterprise 1)

  • Dhevin De Silva
  • Keani Schuller
  • Olivier Larochelle
  • Tiffany Lagarde
  • Valentin Najean

Team Members (Enterprise 2)

  • Vincent El-Ghoubaira
  • Seunghyun Park
  • Olivier Larochelle
  • Tiffany Lagarde
  • Valentin Najean

Project Overview

This project aims to enhance the effectiveness of direct marketing campaigns for banks using machine learning. By leveraging a dataset from the University of California Irvine’s Machine Learning Repository, we explored innovative strategies to attract new clients and retain existing ones, transitioning from traditional to data-driven marketing approaches.

Model Development Insights

  • Classification Models: These models predict potential subscriber behavior, enhancing personalization of marketing campaigns. Techniques include feature selection using Random Forest and performance evaluation through accuracy and F1 score metrics.
  • Clustering Models: Various techniques were assessed, with the best performers identified based on silhouette, Calinski-Harabasz, and Davies-Bouldin scores. This approach segments customers by behavior for targeted marketing strategies.

Advanced Techniques

  • Causal Inference: Introduced to understand the impact of marketing actions, such as campaign frequency and interest rates, on customer decisions.
  • Model Explainability: Tools like SHAP and LIME provide insights into factors influencing model predictions, offering transparency in machine learning processes.
  • Hyperparameter Tuning and AutoML: Techniques like Bayesian optimization and TPOT optimize models for peak performance.

Production and Deployment

  • MLflow: Manages the entire machine learning lifecycle, including experimentation, model tuning, and deployment. It tracks experiments to streamline training, parameter tuning, and evaluation processes.
  • Docker: Used for production, Docker containers ensure that models run consistently across different systems, simplifying deployment cycles and enhancing operational reliability.

Interactive Application

An interactive Streamlit application allows real-time interaction with the model's capabilities, accessible here: https://ui-classfication.azurewebsites.net/