Title

Predictive Analytics for Customer Retention: A Business Intelligence Approach

Creator

Mavroudis, A.

Abstract

Predictive analytics have become an essential component of modern Business Intelligence (BI), enabling organizations to anticipate customer behavior and make data‑driven strategic decisions. This study examines the application of three widely used machine learning algorithms, Support Vector Machine (SVM), Random Forest, and XGBoost, for customer churn prediction using the Customer Churn Prediction Dataset published on Kaggle. The methodology includes algorithm selection based on prior research, preprocessing of heterogeneous customer data, and comparative evaluation of model performance through accuracy, F1‑score, and ROC‑AUC metrics. The results demonstrate that XGBoost outperforms the other models, achieving the highest predictive accuracy and strongest minority‑class performance, making it particularly suitable for churn‑focused BI applications. Random Forest also performs robustly, while SVM shows limitations in handling class imbalance. The discussion highlights the algorithmic mechanisms that explain these differences and emphasizes the business relevance of accurate churn prediction. The paper concludes by outlining key limitations, such as class imbalance, static data, and lack of real‑time integration, and proposes future research directions including hybrid ensembles, deep learning architectures, and real‑time predictive analytics. Overall, the study demonstrates how predictive algorithms can be effectively integrated into BI systems to support proactive customer‑retention strategies and enhance organizational decision‑making.

Keywords

bi

analytics

machine learning

Date Available

May 27, 2026

Type

Poster

Course

CS4848

Instructor

Lazarou, Vassilios

Session

Poster Session

Location

Student Lounge - Main Corridor

Meet the Artist Time

May 27: 12.00 - 15.00

Event

Year 2026 | 10th Annual Student Research and Creative Arts Symposium