Enhancing Retention Strategies through Deep Learning-Based Dropout Prediction

Employees are an organization’s most significant resource. Employee dropout may be costly for firms owing to the costs of recruiting, training, and lost productivity. By forecasting dropout, firms may take preventative actions such as developing retention programs, providing targeted assistance to at-risk employees, and addressing possible workplace concerns. The unpredictable dropout can assist in reducing dropout rates and saving money. Existing approaches to predicting employee dropout use machine learning (ML) techniques for employee dropout prediction, which do not present the correlation of various employee attributes that may have caused the dropout. Moreover, the imbalanced dataset affects the accuracy of prediction results. In this paper, Synthetic Minority Oversampling Technique (SMOTE) is applied to the dataset to solve the issue of imbalanced data. Following that, a deep learning technique, gated recurrent unit (GRU), is utilized to predict staff dropout effectively. It also aided in determining most of the relevant factors of employee results. For this purpose, the IBM employee dataset is utilized for training and assessing GRU using 10-fold test-train splitting. The ultimate objective is to effectively detect dropouts to assist any organization in improving various retention strategies. According to the results, the suggested technique achieves 95% accuracy, more significant than existing state-of-the-art approaches.