Explainable TabNet model predicts workforce performance scores in benchmark dataset

Researchers have built an explainable deep learning framework, called X-TabNet, that predicts employee performance scores from multidimensional workplace data, reporting very high accuracy on a benchmark dataset of 100,000 workforce records. The study, published in Scientific Reports, was led by Zihe Zhu, Guangsen Wei, XiaoQian Bian and Sihui Yu, and was released on 25 October 2026 as an open-access article.

The framework builds on TabNet and adds explainability and fairness evaluations on top of the prediction task. In the main evaluation, TabNet achieved an R² of 0.9880, a mean absolute error of 0.1223 and a root mean square error of 0.1549. Five-fold cross-validation produced a mean R² of 0.9723, with a standard deviation of 0.0201 across the folds.

The team also tested robustness under more demanding splitting schemes. With repeated random splits the model reached an R² of 0.9618, plus or minus 0.0087. Group-aware splitting yielded an R² of 0.9483, chronological splitting an R² of 0.9326, and out-of-distribution evaluation an R² of 0.8874. The declining scores across these harder tests show that performance drops as the evaluation conditions become more realistic, with the largest gap appearing when the model is applied to data distributions it has not seen during training.

In addition to the regression task, the researchers formulated the prediction problem as an ordinal classification task. This ordinal formulation reached an accuracy of 94.18 percent and a Macro-F1 score of 0.9257. The framework also included full leakage checks, feature importance analysis, explainability evaluation and subgroup fairness assessment, and the authors report that differences in subgroup errors across the assessed demographic groups were generally small.

The authors are explicit that the results should not be overinterpreted. Owing to the provenance and representativeness of the dataset, they state that it is far too early to draw conclusions regarding the relationship between the model’s predictions and real organizational environments. They note that additional independent validation with organizational data is required before any operational use of the framework.

Within those limits, the study presents X-TabNet as a methodological framework for tabular workforce-related prediction problems rather than as a deployable management tool.

Subject of Research: Explainable machine learning for employee performance prediction from tabular workplace data

Article Title: An explainable and fair TabNet framework for remote employee performance prediction using multidimensional workplace data

Article References: Zhu, Z., Wei, G., Bian, X., & Yu, S. (2026). An explainable and fair TabNet framework for remote employee performance prediction using multidimensional workplace data. Scientific Reports. https://doi.org/10.1038/s41598-026-76297-z

Image Credits: AI Generated

DOI: 10.1038/s41598-026-76297-z

Keywords: TabNet, machine learning, employee performance, remote work, explainable AI, algorithmic fairness, tabular data, regression, cross-validation, Scientific Reports, explainable, fair