The problem
Predict which customers are likely to buy caravan insurance. It's a classic imbalanced marketing-response problem: buyers are a small minority of the base, so the data has to be balanced before any model will behave, and the real prize is knowing who to target rather than chasing everyone.
Approach
Ran the full KDD (Knowledge Discovery in Databases) process, objectives, data exploration, preprocessing, selection, modelling, evaluation. Balanced the classes, used recursive feature elimination to cut the feature set to what mattered, and compared six models, logistic regression, random tree, neural network, CHAID, random forest and decision tree, across SPSS and R. Built an interactive Tableau dashboard to communicate the findings.
Result
A champion model chosen on the ratios, and four named customer segments (Young Professionals, Family-Oriented, Retirees and Urban Singles), each with its traits, needs and a recommended marketing approach. That final step, turning a model score into segments a marketing team can actually act on, is the point of the whole exercise.
Tools
SPSS R Tableau KDD
