Employee Attrition Predictive Modeling Using CRISP-DM: A Supervised Learning and Workforce Analytics Framework
Transactions on Engineering and Computing Sciences · 20.11.2026 · 14 (06)
DOI: 10.14738/tecs.1406.12677 ↗Kurz gefasst
Fachartikel in der Zeitschrift Transactions on Engineering and Computing Sciences, online erschienen am 20.11.2026. Verfasst von Upendra Kanuru und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Employee turnover represents a critical operational, financial, and strategic bottleneck across contemporary enterprise environments, frequently incurring replacement expenditures ranging between 50% and 200% of an annual salary while eroding institutional human capital. This study presents a rigorous end-to-end predictive modeling framework adhering strictly to the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology to identify empirical risk factors and reliably forecast voluntary employee turnover prior to formal resignation. Leveraging a curated enterprise human resources dataset encompassing 1,270 personnel records across 35 multi-modal attributes, rigorous data sanitation, automated zero-variance filtration, and factor encoding were performed alongside Exploratory Data Analysis, Recursive Partitioning (Decision Trees), Principal Component Analysis (PCA), and Unsupervised K-Means clustering ( ). Supervised predictive models—specifically Binomial Logistic Regression and Naïve Bayes Classification—were trained and benchmarked against an independent 30% test holdout ( ) and further validated against a 200-record unseen verification holdout. The primary institutional research goal was to determine whether voluntary attrition could be classified with an accuracy exceeding 70.0%. Experimental results demonstrate that Logistic Regression substantially outperformed Naïve Bayes, achieving a test accuracy of 88.71% (95% CI: 85.10%–91.71%), an Area Under the ROC Curve (AUC) of 0.8282, a sensitivity of 98.11%, and a Cohen’s Kappa of 0.4906. When evaluated against the unseen verification cohort, the model demonstrated outstanding generalization capability with an accuracy of 91.00%, a sensitivity of 98.80%, an improved specificity of 51.52%, and a Kappa of 0.6064. Key empirical predictors of attrition identified include overtime commitments, business travel frequency, tenure without promotion, distance from home, and reported environmental satisfaction. A modular production turnover pipeline and ethical governance framework are formalized to operationalize real-time individual risk stratification while safeguarding against algorithmic bias.
Zitieren
APA
Kanuru, U., Duvvada, V. C., & Cheepuri, S. (2026). Employee Attrition Predictive Modeling Using CRISP-DM: A Supervised Learning and Workforce Analytics Framework. Transactions on Engineering and Computing Sciences, 14(06). https://doi.org/10.14738/tecs.1406.12677
Vancouver
Kanuru U, Duvvada VC, Cheepuri S. Employee Attrition Predictive Modeling Using CRISP-DM: A Supervised Learning and Workforce Analytics Framework. Transactions on Engineering and Computing Sciences. 2026;14(06). doi:10.14738/tecs.1406.12677
BibTeX
@article{Kanuru2026,
author = {Kanuru, Upendra and Duvvada, Vinay Chowdary and Cheepuri, Sirisha},
title = {Employee Attrition Predictive Modeling Using CRISP-DM: A Supervised Learning and Workforce Analytics Framework},
journal = {Transactions on Engineering and Computing Sciences},
year = {2026},
volume = {14},
number = {06},
doi = {10.14738/tecs.1406.12677}
}
EUVIMED