Research on deep learning-based methods for predicting employee turnover in enterprises

In managing employee turnover, companies urgently need to leverage scientific forecasting methods to support decisionmaking, thereby maximizing the retention of key talent. Although existing methods have made progress in this field, issues such as training instability, insufficient representation of feature interactions, and class imbalance persist, limiting the accuracy and robustness of predictions. To this end, this paper proposes the CO-ResMLP employee turnover prediction model, which integrates residual structures, coordinate attention, and focus loss. This model enhances training stability and depth scalability through using residual blocks, introduces coordinate attention to improve feature interaction modelling capabilities, and employs focal loss to mitigate class imbalance. Experimental results on the IBM dataset demonstrate that the model achieves an accuracy of 0.947, a precision of 0.944, a recall of 0.922, an F1-score of 0.931, and an AUC of 0.986. These metrics represent improvements of 0.031, 0.022, 0.066, 0.045, and 0.025, respectively, over traditional models. The results demonstrate that this model can more accurately identify employees at high risk of turnover. While maintaining a high recall rate, it significantly improves precision, reduces false positives, and provides reliable decision support for human resource management.