Bank Marketing - Data Mining Project 🏦📊

📌 Project Overview

This repository contains a semester project for the "Statistical Methods of Data Mining" course at the Czech University of Life Sciences Prague (ČZU v Praze).

The primary goal of this project is to analyze a real-world dataset from a real Portuguese bank, sourced from Kaggle. We built predictive models to determine whether a client will subscribe to a term deposit (target variable: y) based on the results of their telemarketing campaigns.

Authors: Ondřej Kotala, Daiana Bikmaeva, Robert Sládek, Pavel Dikan
Year: 2025

🛠 Tools & Techniques

• Tool: IBM SPSS Modeler
• Techniques Used: * Data Understanding & Exploration
  • Data Preparation (Handling missing values, outlier treatment)
  • Feature Engineering (Addressing multicollinearity, capping extreme values)
  • Predictive Modeling & Evaluation

📂 Repository Structure

• /data/ — Contains the campaign datasets (bank-additional-full.csv, bank-additional.csv) and the variable description text file.
• /models/ — Contains IBM SPSS Modeler streams (.str) and generated models (.gen).
• /docs/ — Comprehensive project documentation and reports (e.g., MARKETING V5_1.pdf).

📊 Key Insights & Data Preparation

During the Data Understanding and Data Preparation phases, several critical steps were taken to ensure high model quality:

• Hidden Missing Values: Addressed implicit missing data (labeled as "unknown") across categorical variables.
• Multicollinearity: Identified and resolved high correlation among macroeconomic indicators (euribor3m, emp.var.rate, and nr.employed) by keeping only the most relevant variable.
• Outlier Capping: Handled extreme values in variables like campaign (number of contacts). The right tail was capped at the 99th percentile (value of 14) to make the distribution more compact and robust for modeling.
• Target Leakage: Removed the duration variable as highly predictive but unknown before a call is actually made, ensuring a realistic predictive model.

📄 Documentation

For a deep dive into the business logic, methodology, statistical analysis, and final model evaluation, please refer to the main documentation file located in the /docs/ folder.

🎓 What I Learned (Skills Acquired)

• End-to-End Analytics: Mastered the complete data mining lifecycle using the CRISP-DM framework, from business understanding to model evaluation.
• Data Quality Management: Learned how to identify and solve critical data issues in real-world datasets, such as hidden missing data (e.g., "unknown" categories), outliers, and target leakage.
• Statistical Rigor: Gained practical experience in exploratory data analysis (EDA), correlation analysis, and resolving multicollinearity to build simpler, more reliable predictive models.
• Visual ML Pipelines: Developed strong hands-on skills in creating visual machine learning workflows and building binary classification models using IBM SPSS Modeler.