ORIGINAL RESEARCH article
Front. Educ.
Sec. Higher Education
KASTLE-X: Incremental Value of Kolmogorov–Arnold Network Augmentation in a Stacked Tabular Ensemble for Employee Attrition Prediction
- NP
Nirodh Paul 1
- MH
Md. Mahedi Hassan 2
- SF
Sumnoon Ferdous 3
- MH
Md. Masuqul Haque 4
Mir Mohammad Azad 5
1. Department of Accounting and Information Systems, University of Barishal, Barishal, Bangladesh
2. Department of Software Engineering, Daffodil International University, Dhaka, Bangladesh
3. Department of Management Information Systems, Independent University Bangladesh, Dhaka, Bangladesh
4. Department of Chemistry, University of Rajshahi, Rajshahi-6205, Bangladesh
5. Department of Computer Science and Engineering, Uttara University, Turag, Uttara, Dhaka-1230, Bangladesh
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Abstract
Employee attrition prediction from routine personnel records is a canonical imbalanced tabular classification problem. We propose KASTLE-X (Kolmogorov–Arnold Network-Augmented Stacked Learning Ensemble with leakage-free feature engineering) and investigate whether an architecturally novel KAN component adds incremental value to a conventional stacking ensemble of Bayesian-optimized tabular learners. The framework is evaluated on the IBM HR Analytics Employee Attrition benchmark (1{,}470 records, 16.12% positive prevalence) under a nested 5{×}5-fold stratified cross-validation protocol in which all preprocessing, hyperparameter optimization, meta-feature generation, calibration, and threshold selection are confined strictly to training partitions. We first benchmark 20 individual classifiers spanning eight algorithmic families; KAN-Bag, a seed-bagged Chebyshev-polynomial KAN, achieves the highest individual AUC (0.8485 ± 0.0266) and F1 (0.5945). We then establish that adding KAN-Bag to a seven-learner conventional stacking pool improves ensemble AUC by +0.0114 (p = 0.0036, paired t-test; Wilcoxon minimum p = 0.0625, n=5 folds), PR-AUC by +0.0188, and F1 by +0.0698 on every outer fold. The logistic meta-learner assigns KAN-Bag the largest coefficient (2.51), confirming its contribution arises from architectural diversity rather than raw accuracy alone. We contextualize the 0.0114 AUC gain—achieved with only 237 positive cases—against its practical screening benefit (3.7× enrichment over base rate) and against the power limits of five-fold testing, and we clearly bound all claims to this single synthetic benchmark. These results support the narrower but well-grounded conclusion that Kolmogorov–Arnold networks provide distinct diversity that may benefit tabular stacking ensembles under class imbalance, with external validation on other organizational datasets as a necessary future step.
Summary
Keywords
employee attrition; Kolmogorov–Arnold networks; stacked ensemble learning; feature engineering; class imbalance; nested cross-validation; explainable AI; Bayesian optimization; tabular deep learning
Received
20 October 2026
Accepted
01 November 2026
Copyright
© 2026 Paul, Hassan, Ferdous, Haque and Azad. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Mir Mohammad Azad
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.