Chapter 7

Title: Who Wants My Product? Affinity-Based Marketing

Summary

Chapter 7 addresses the task of product affinity-based marketing and optimizing a direct marketing campaign. A bank has introduced a new financial product, a new type of current (checking) account, and some of its customers have already opened accounts of the new type, but many others have not done so yet. The bank’s marketing department wants to push sales of the new account by sending direct mail to customers who have not yet opted for it. However, in order not to waste efforts on customers who are unlikely to buy, they would like to address only those customers with the highest affinity for the new product. Binary classification is used to predict for each customer, whether they will buy the product, along with a confidence value indicating how likely each of them is to buy the new product. Customers are then ranked by this confidence value and the 20% with the highest expected probability to buy the product are chosen for the campaign.

Following the CRoss-Industry Standard Process for Data Mining (CRISP-DM) covering all steps from business understanding and data understanding via data pre-processing and modeling to performance evaluation and deployment, this chapter first describes the task, the available data, how to extract characteristic customer properties from the customer data, their products and accounts data and their transactions, which data pre-processing to apply to balance classes and aggregate information from a customer’s accounts and transactions into attributes for comparing customers, modeling with binary classification, evaluating the predictive accuracy of the model, visualizing the performance of the model using Lift charts and ROC charts, and finally ranking customers by the predicted confidence for a purchase to select the best candidates for the campaign. The predictive accuracy of several learning algorithms including Decision Trees, Linear Regression, and Logistic Regression is compared and visualized comparing their ROC charts. Automated attribute weight and parameter optimizations are deployed to maximize the prediction accuracy and thereby the customer response, sales volume, and profitability of the campaign. Similar processes can be used for customer churn prediction and addressing the customers predicted to churn in a campaign with special offers trying to prevent them from churning.

Table of Contents

7.1 Introduction
7.2 Business Understanding
7.3 Data Understanding
7.4 Data Preparation
7.4.1 Assembling the Data
7.4.2 Preparing for Data Mining
7.5 Modelling and Evaluation
7.5.1 Continuous Evaluation and Cross Validation
7.5.2 Class Imbalance
7.5.3 Simple Model Evaluation
7.5.4 Confidence Values, ROC, and Lift Charts
7.5.5 Trying Different Models
7.6 Deployment
7.7 Conclusions
7.7 Glossary
7.7 Bibliography

Dataset & Processes: Click here to download