Improved Adam Optimizer with Warm-Up Strategy and Hyperbolic Tangent Function for Employee Turnover Prediction Using Multilayer Perceptron (MLP)

Employee turnover poses significant challenges for organizations, affecting both operational efficiency and financial stability. Accurately predicting employee turnover can help organizations implement proactive measures to retain valuable talent. This research investigates an enhancement to the Adam optimizer by integrating a warm-up strategy with the Hyperbolic Tangent (tanh) function, aimed at reducing the number of iterations or epochs needed to train Multilayer Perceptron (MLP) models for predicting employee turnover. The Adam optimizer, known for its adaptive learning rate and computational efficiency, can require extensive training periods to reach optimal performance. Incorporating a tanh-based warm-up phase improves the stability and convergence speed of the training process. This study shows that the proposed approach not only reduces training duration but also enhances the predictive accuracy of MLP models. The results indicate that the Adam optimizer with the tanh-based warm-up strategy significantly outperforms other optimizers, including Nadam, AdaMax, and AdamP, achieving the highest accuracy (95%) and it required the fewest iterations to converge (50 epochs. These findings suggest that this novel optimization technique can be valuable for organizations seeking efficient and accurate employee turnover prediction models. Future research should explore the applicability of this approach across different datasets and neural network architectures to further validate these findings.