Employee Attrition Prediction Using Temporal Convolutional Network (TCN) on HR Time Series Dataset to Support Employee Retention Strategies

Abstract

Abstract— Employee attrition continues to be a major challenge for organizations because high turnover increases recruitment costs, disrupts business operations, and affects workforce stability. Predictive analytics provides an opportunity to identify employees who are at risk of leaving before turnover occurs. A Human Resources dataset collected between 2018 and 2024 served as the basis for evaluating the Temporal Convolutional Network (TCN) in employee attrition prediction. The dataset contains 16,050 employee records described by 25 attributes covering demographic information, job characteristics, compensation, and work experience. Data preprocessing included feature selection, categorical encoding, normalization, and sequence preparation before model training. The proposed model was evaluated using three train–validation–test splitting strategies and assessed with Accuracy, Precision, Recall, F1-score, and Area Under the ROC Curve (AUC). Experimental results show that the TCN model consistently achieved high predictive performance across different experimental settings, with the best configuration obtaining an Accuracy of 98.35%, an F1-score of 0.8976, and an AUC of 0.9819. These findings demonstrate that TCN provides stable and reliable employee attrition prediction and has the potential to support Human Resource Analytics through earlier identification of employees with a higher risk of leaving the organization.

Keywords—Employee Attrition, Temporal Convolutional Network, Deep Learning, Human Resource Analytics, Employee Retention.