Experimental Evaluation of a Novel Methodology to Predict Employee Attrition Using Elevated Deep Gradient Optimization Logic

Employee attrition prediction is a critical challenge for organizations as it impacts productivity, continuity, and financial performance. This study proposes a novel cascaded methodology combining Multi-Layer Perceptron (MLP) and Support Vector Machine (SVM) models, enhanced by deep gradient optimization techniques, to predict employee attrition. The dataset, preprocessed through feature selection and synthetic oversampling (SMOTE), included key employee features such as monthly income, job satisfaction, and overtime. The proposed methodology was tested against other models, including Random Forest (RF) and Logistic Regression (LR). The cascaded MLP+SVM model achieved a training accuracy of 94.5% and a test accuracy of 92.1%, outperforming standalone MLP (89.7%) and SVM (86.5<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>). The model also showed minimal overfitting, with only a 2.4 % difference between training and test accuracy. Key performance metrics, including an F1 score of 0.83 and an AUC score of 0.91, confirmed the model's ability to classify employees likely to leave accurately. The model's ability to predict employee attrition helps organizations proactively retain critical talent by addressing factors like job dissatisfaction and high overtime.