Here I am using a sample dataset from IBM Watson Analytics to replicate the Churn Modelling project I did at my previous employer, Claro. The dataset has a Class-Imbalance problem, and I solve it by Under Sampling the dominant class. I perform an exploratory analysis in order to identify meaningful predictors of churn, and then use a Logistic Regression to predict who churned and who didn’t. Finally, I evaluate the model’s performance with by cross-validating with a hold-out sample.
This is more of an academic project. I am working with a former colleague to publish a paper discussing something we found using internal data when I was working at Claro. Customer Satisfaction Indices – including the infamous NPS (Net Promoter Score) – should not be taken at face value without first correcting for several biases; amongst them is the effect of weather patterns across time.