Year: 2026
Type: article
Abstract: This paper investigates the effectiveness of feature selection techniques in optimizing supervised machine learning pipelines for customer churn prediction using the publicly available Customer Churn Dataset from Kaggle. Feature selection plays a crucial role in enhancing model interpretability and generalization by eliminating irrelevant or redundant variables. In this study, three optimized feature selection algorithms—Minimum Redundancy-Maximum Relevance (mRMR), Multi Spatially Uniform Relief (MultiSURF), and Hilbert-Schmidt Independence Criterion (HSIC)—were implemented and compared alongside a baseline scenario without feature selection. Eight supervised machine learning classifiers, including Support Vector Machine (SVM), Logistic Regression, K-Nearest Neighbors (KNN), Decision Tree, AdaBoost, Bagging, Stacking, and Voting Classifier, were optimized using Randomized GridSearchCV. The dataset was divided into training and testing subsets with a 70:30 ratio, and model performance was evaluated using metrics such as accuracy, precision, recall, F1-score, ROC-AUC, and computational time. Experimental results reveal that all feature selection techniques achieved comparable performance across classifiers. This study concludes that optimized feature selection techniques contribute to more stable and efficient model training.
Author: Merit Chinonso Opara
Institution: Madonna University
Language: English
Subfield: Marketing
Domain: Social Sciences
Open Access status: bronze
OpenAlex ID: W7213282945
