An issue for every Company is the high number of churning customers. Company's customers lapse or quit insurance policies every day. Some of them for unpredictable reasons but others for the competitive environment. So, many customers leave insurance policies for reasons that a Company can prevent with the help of the churn prediction.
In this work, from the collection of portfolio contracts, will be predicted the probability of lapses, with the both use of traditional and modern machine learning, also facing issues linked with imbalanced classes.
The growing of open source programming communities combined with the advances in computing power have given the opportunity to develop and use models with more predictive power than traditional ones. Anyway, ensemble algorithms and neural networks are complex models referred as "black-boxes" than linear regression, decision trees and logistic regression referred as interpretable models. For this reason, it's becoming more common to combine the predictive power with a toolbox to understand results from machine learning.
