Modeling pedestrian emotion in high-density cities using visual exposure and machine learning: Tracking real-time physiology and psychology in Hong Kong

Abstract

People's mental health has been deteriorating as a result of urban living, which also causes disability, pain, or even death among metropolitan residents. Understanding the interaction between human emotions and the built environment is therefore essential for developing strategies towards a psychologically friendly and socially resilient environment. This study aims to model the pedestrian emotion in high-density urban areas of Hong Kong using machine learning based on environmental visual exposure, which is one of the most significant factors affecting emotions. Using ambulatory sensing and portable technologies, two-dimensional emotional data from 99 pedestrians are retrieved from coupled data with respect to wearable arousal, subjective report preference, and the location at a high spatiotemporal resolution (4 m). The visual environment is quantified by isovist and street view factors. This study examines the impact of visual exposures using two multiple linear regression (MLR) models and establishes a predictive model using random forest (RF) (N = 548). The MLR models have R2 values around 0.5 and suggest that in the high-density environment, exposure to more trees, visual volume, and drift magnitude can cause positive emotions. Conversely, areas with a view of sign symbols, object proportion, min-radial, and occlusivity can cause negative emotions. The resultant predictive model with nine visual exposure variables can explain 79% of the spatial variance of pedestrian emotion. Furthermore, the methodological framework provides opportunities for spatial, data-driven approaches to portable sensing and urban planning research. The findings are also applicable to optimize the infrastructure design of outdoor environments for more psychologically friendly experiences.