Even in these difficult times, we are committed to our students and their work. In this context, I want to announce that Chen Ofir had successfully defended his M.A. project “Analysing Feature Selection for Churn Prediction Models: Exploring the Impact of ML Model Building Blocks”.
Full report can be accessed here.
In today's fiercely competitive business landscape, customer churn poses a significant challenge for organizations across various industries. Accurately predicting and mitigating customer churn has become crucial for businesses striving to maintain profitability and foster long-term customer relationships. Machine learning (ML) models, such as Optimove's CatBoost, have emerged as valuable tools for churn prediction due to their ability to leverage vast amounts of customer data and identify patterns indicative of churn behavior. However, the black-box nature of these models can limit our understanding of the features that contribute most significantly to accurate predictions, therefore when trying to improve models the data teams have no lead on what type of data points have better potential of leading to model improvement. In this research we analyzed 85 Catboost models generated by Optimove for Churn prediction and aim to explore the impact of features selected by the model and their effectiveness. By analyzing the lift in churn prediction for Catboost models generated on different data sets from the marketing industry, we seek to uncover insights into the specific attribute types that enhance the model's predictive capabilities. Understanding the role of these features can enable businesses to make informed decisions for developing and exposing additional data points or improving customer-focused strategies, ultimately leading to better churn prediction and retention efforts.