A machine learning based pipeline for predicting performance of call center operators

Predicting employee performance is critical for strategic human resource planning, as organizational success heavily relies on workforce productivity and attendance. This study evaluates machine learning techniques to forecast absenteeism and productivity among call center operators, leveraging a real-world dataset of approximately 10,500 employees from a large Brazilian call center. The dataset, derived from employee profiles, underwent preprocessing, including correlation-based filtering and feature selection, to identify the most predictive variables. We compared the performance of five machine learning models — Logistic Regression (LR), Multi-layer Perceptron (MLP), Naive Bayes (NB), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) — using a dynamic pipeline that integrates evolutionary algorithms for hyperparameter tuning and feature selection. The models were evaluated using nested cross-validation to ensure robustness. Results demonstrate strong predictive performance, with LR achieving the highest ROC AUC scores for absenteeism and productivity: 0.78 and 0.87, statistically similar with the results obtained by XGBoost (0.74 and 0.86) and RF (0.76 and 0.86). These findings highlight the effectiveness of machine learning in optimizing workforce management, particularly in high-turnover environments like call centers. The study also emphasizes the practical applicability of the proposed pipeline, which has been successfully integrated into the company’s management system, yielding measurable improvements in operational efficiency. This work contributes to both academic research and industry practice by providing a scalable, data-driven approach to performance prediction while adhering to ethical and regulatory standards in employee data usage.