A transformer-based deep learning framework to predict employee attrition.

Abstract

In all areas of business, employee attrition has a detrimental impact on the accuracy of profit management. With modern advanced computing technology, it is possible to construct a model for predicting employee attrition to minimize business owners' costs. Despite the reality that these types of models have never been evaluated under real-world conditions, several implementations were developed and applied to the IBM HR Employee Attrition dataset to evaluate how these models may be incorporated into a decision support system and their effect on strategic decisions. In this study, a Transformer-based neural network was implemented and was characterized by contextual embeddings adapting to tubular data as a computational technique for determining employee turnover. Experimental outcomes showed that this model had significantly improved prediction efficiency compared to other state-of-the-art models. In addition, this study pointed out that deep learning, in general, and Transformer-based networks, in particular, are promising for dealing with tabular and unbalanced data.