AI-Driven Predictive Analysis for Customer Retention Prediction
5/13/2024
Authors: Ing. Ilham Supriyanto and Elif Yozkan, Advantech Europe
In the modern landscape of hybrid businesses, where both B2B and B2C relationships intertwine, accurately predicting customer churn is essential. This study presents a predictive framework using real-world data, employing advanced Deep Neural Network (NN) and Gradient Boosting (XGBoost) models for an Industrial Computer Business. Our research indicates that the NN approach outperforms XGBoost, as evidenced by the Euclidean distance metric analysis comparing misclassified instances, specifically by assessing the similarity between incorrect predictions and the actual outcomes.
To address the issue of unequal representation of different classes in our data, which could lead to biased predictions, we implemented a technique called under-sampling with Tomek Links, that aims to address the issue of class imbalance by removing certain samples from the majority class. Our models achieved impressive performance with an average F1 score of 0.85 out of 1 across multiple years, exceeding the industry-standard accuracy threshold of 0.80. Furthermore, we illustrate a deployment strategy integrating Robotic Process Automation (RPA) and AI, showcasing their cooperation in digital solution development and providing practical insights for deploying predictive models in dynamic business settings.
A presentation on this topic will be conferred at the IntelliSys 2024 (Intelligent Systems Conference) on September 5 & 6 2024 in Amsterdam, Van der Valk Hotel Amsterdam Amstel, The Netherlands
