Improving employee performance rating prediction using ANN probability-enhanced ensemble learning

Employee performance rating prediction is an important task in human resource management. It helps organizations identify high-performing employees, improve training programs, and support decisions regarding promotions and workforce allocation. However, employee performance is influenced by multiple factors such as job position, work experience, job satisfaction, overtime, department, and managerial environment. These factors are typically nonlinear and interrelated, making accurate prediction challenging. To address this issue, this study proposes an ANN probability-enhanced ensemble learning framework, in which the soft class-probability outputs generated by multiple ANN base learners are used for both weighted prediction fusion and feature enhancement by combining them with the original employee features for second-stage classification. First, multiple ANN base learners with different hidden layer structures are trained to obtain diverse predictive information. Then, the output probabilities of these base learners are utilized in two ways: weighted ANN ensemble and probability-enhanced feature construction. The probability-enhanced features are further combined with various meta-classifiers, including Random Forest, ExtraTrees, Decision Tree, K-Nearest Neighbors, HistGBDT, AdaBoost, Logistic Regression, and Linear Support Vector Machine (LinearSVM). Experiments were conducted on the IBM HR Analytics Employee Attrition and Performance dataset from Kaggle. Results show that the proposed ANNProb-AdaBoost model achieves the best macro F1 score of 0.4885 and an accuracy of 0.8265. Compared to individual ANN models and traditional machine learning models, this approach delivers more balanced classification performance. Additionally, sensitivity analysis and feature importance analysis based on feature deletion were performed to evaluate the model's robustness and interpretability.