Comparative Analysis of High- and Low-Performing Factory Workers with Attention-Based Neural Networks

This study presents a new method that supports the comparative analysis of works performed by high- and low-performing factory workers. Our method, based on explainable deep learning, automatically detects a sensor data segment that potentially contains knowledge about the skill of works by analyzing acceleration sensor data from high- and low-performing workers. Our evaluation with industrial engineers using sensor data from actual factory workers revealed that 78% of sensor data segments detected by our method included knowledge about skill.