Machine Learning-Based Prediction of Employee Performance Using Lifestyle and Diet Indicators

Abstract

This study presents a machine learning-based framework to predict daily work productivity based on nutrition and lifestyle data collected from 310 Vietnamese employees. Four 4,000 daily observations were used to train and evaluate three models: Random Forest, XGBoost, and Multi-layer Perceptron (MLP). Among them, the XGBoost model optimised with Optuna achieved the best performance with RMSE = 2.9, MAE = 2.5, and R² ≈ 0.92. Feature importance analysis revealed that protein intake, caloric consumption, and physical activity were the most influential predictors. The model was also evaluated as a classifier, achieving over 83% accuracy and a macro F1-score above 0.80. These findings demonstrate the feasibility of integrating AI and personalised nutrition strategies to improve workforce productivity in real-world settings. The proposed system offers potential for use in corporate wellness programmes and digital health platforms in developing countries.