TRANSFER LEARNING IN VOLUNTARY TURNOVER PREDICTION MODELS

This work investigates the transfer learning in voluntary turnover prediction methods. Many companies, especially small and medium-sized ones, do not have enough data to train predictive models, making the use of pre-trained models essential. We investigated the performance of the two most used models in the literature for this problem: logistic regression (LR) and random forests (RF). The models were trained and tested on two public datasets from different sources. The results show that the RF classifier achieves higher accuracy than LR in all scenarios. In the study on transfer learning, the model trained on dataset 1 had a good ability to predict turnover events in dataset 2; however, the second model, trained on dataset 2, performed poorly when tested on dataset 1. This suggests that transfer learning may depend on the training context, indicating the need for diversity in the data used to train the model.