Employee tardiness prediction using machine learning and large language models: a case study in logistics enterprises

Employee tardiness poses a significant management challenge for enterprises, particularly in logistics and e-commerce sectors where operational efficiency is paramount. This study proposes a hybrid artificial intelligence framework integrating traditional machine learning models with open-source large language models (LLMs) to predict the likelihood of employee tardiness. By incorporating actual attendance data from JD Group, the research synthesises multisource features including historical clock-in records, weather conditions, commuting distance, and personal attributes. Performance comparisons were conducted between six classical machine learning classifiers-logistic regression, stochastic gradient descent, naive Bayes, support vector machines, random forests, and k-nearest neighbors-and three fine-tuned LLMs (Qwen-7B, ChatGLM-6B, and Yi-6B). Experimental results indicate that the Qwen-7B model achieves the highest prediction accuracy at 89.85%, with precision, recall, and F1 scores all exceeding 89.5%. The study details technical implementation aspects, including the LoRA (Low-Rank Adaptation) fine-tuning method and a comprehensive data preprocessing workflow. Furthermore, this study introduces a cost-benefit analysis framework based on JD Group's internal research into proxy clocking, revealing that the proposed AI-driven approach can significantly reduce labour cost losses and enhance management efficiency. This research provides a feasible, scalable solution for intelligent human resource management in data-rich corporate environments, advancing the application of computer vision and machine learning in industrial settings.