Classifier Optimization with a Test-Time Training Layer for Employee Attrition Prediction

Employee attrition poses significant challenges for organizations, impacting productivity and increasing recruitment costs. This paper presents an innovative approach to optimize baseline classifiers by integrating a Test-Time Training (TTT) layer. The proposed method refines model predictions on unseen data, enhancing adaptability to new patterns in employee behavior. We evaluate the performance of the optimized model using the IBM Analytics HR Employee Attrition dataset, which includes 1,470 observations and 34 features. The dataset was split into training and test sets, with 80% of the data used for training and 20% for testing. The results demonstrate that incorporating a TTT layer significantly improves prediction accuracy compared to the baseline Logistic Regression model, the SVM model and the Random Forest model with notable enhancements in metrics such as accuracy, precision, and F1-score. These findings suggest that this approach provides organizations with a robust tool for predicting employee attrition.