EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2137. Interpretable Ensemble Learners for Explainable Business Analytics

Invited abstract in session WC-27: Machine Learning and Ensemble Learning with optimization methods, stream Mathematical Optimization for XAI.

Wednesday, 12:30-14:00
Room: 047 (building: 208)

Authors (first author is the speaker)

1. Eren Berk Aytac
2. Koen W. De Bock
Department of Marketing, Audencia Business School
3. Sureyya Ozogur-Akyuz
Department of Mathematics Engineering, Bahcesehir University

Abstract

Customer churn is a common problem for companies. The problem is that traditional business analytics models, particularly when it comes to predicting customer churn. The core of the issue is that these traditional models don't necessarily align with business objectives.
For instance, a typical customer churn model might tell us whether a customer is likely to discontinue service, but it stops short of considering the financial implications of churn. Additionally, algorithms' decision-making processes and outcomes should be understandable and explainable to humans. Rule-based models (e.g.rule ensembles) are among the methods used to provide this level of explainability.
In this study,Profit Driven Spline Rule Ensembles Customer Churn Prediction Model has been developed for this problem. The unique aspect here is the profit-driven approach, which ensures that the predictions are aligned with the company's financial goals. Part of this approach involves adapting model performance metrics EMPC to be understandable to humans. By utilizing spline rule ensembles, the model captures complex relationships in customer behavior that standard models might miss. The key to the model's success lies in its interpretability. In today's complex decision-making landscape, clarity is very important. The model doesn't just churn out predictions; it provides a window into the 'why' and 'how' of customer behavior, empowering decision-makers with the knowledge to act decisively and effectively.

Keywords

Status: accepted

Back to the list of papers