Predictive Model to Predict Donors for their upcoming Yearly campaign and help them have a Targeted Marketing Approach
Targeted Marketing Always Reaps more Profits than Mass Marketing Campaigns. Realizing this strategy Belgium based Donor company approached us to Create a Predictive Model which Could Predict the Donors who would donate > 35 Euros in their next upcoming Reactivation Yearly campaign.
Steps and Algorithms / Techniques used are as follows:
Data Cleaning, Handling Data Quality in a Structured Manner, Amazing Business insights were created from historical Data and fed as variables, Prediction Model was built on R using Undersampling and Mixed Sampling Along with 5*2 Validation and Bagging. Machine Learning, Models were built on Logistic Regression, Decision Trees and Random Forest and based on AUC, LIFT Values one of Model was Baselined
Achieved a Lift Value of 2.03 in top 10%, 2.13 in Top20%, 1.77 in Top 30% which meant they could cover 20% Donor Predicted with just 10% of Customers and 42% of Actual Donors Predicted with just 20% of Customers. Overall 89% Predicted correct from the whole data set of the past. This would save DSC huge letter sending costs as they could focus more on Targeted Marketing Approach.
Interesting Algorithms Applied :
Undersampling + Bagging (U-Bagging) with Logistic Regression Mixed Sampling with Logistic Regression Forward feature selection Calculating performance parameters like AUC, Lift, Sensitivity, etc.