Human-centric lighting intelligence: A data-driven framework for multivariable response analysis and cluster-based prediction

초록

This study aimed to comprehensively investigate the effects of light environments in office spaces on human physical, psychological, and work-related responses. While previous studies relied on limited combinations of illuminance and correlated color temperature (CCT) and focused primarily on mean comparisons and significance testing, this study proposed a machine learning-based analytical framework to integratively interpret human responses. To this end, experiments were conducted with 72 participants under 18 lighting conditions combining six illuminance levels and three CCT levels, resulting in a total of 345 valid datasets. Response indicators included physical (skin conductivity, heart rate variability [HRV]), psychological responses (preference, visual comfort, fatigue), and work performance (accuracy, speed). The analytical framework consisted of (1) data normalization and correlation structure analysis, (2) PCA-based hierarchical clustering, (3) machine learning-based predictive modeling, and (4) interpretation of variable contributions. The results showed that illuminance served as the primary axis distinguishing response types, while CCT acted as a secondary axis inducing finer classification within the same illuminance level. Additionally, HRV indicators (RMSSD, SDNN), visual comfort and preference, accuracy, and speed were key factors. These findings quantitatively elucidate the structural hierarchy of human responses to changes in illuminance and CCT, clarifying the nonlinear characteristics of human responses. Furthermore, this research provides an empirical foundation for data-driven lighting environment design and the development of personalized lighting solutions. Although it can be expanded into precise personalized models through the accumulation of long-term data under diverse lighting conditions in the future.

키워드