Assessing human emotional experience in pedestrian environments using wearable sensing and machine learning with anomaly detection

Enhancing the walkability of pedestrian environments is essential for promoting physical and mental health. Increasing attention has been directed toward the subjective dimensions of walkability, such as individuals’ emotional responses to specific environments, due to their significant association with walking intentions. However, assessing subjective feelings through surveys is challenging to apply consistently across numerous alleyways. Therefore, this study investigates the potential of smart wearable sensors to assess pedestrians’ emotional experiences during walking. Specifically, the research focuses on classifying emotional states, which are categorized as pleasant or unpleasant (i.e., valence)–within pedestrian environments. This classification is achieved by integrating multi-sensor data and anomaly detection techniques. Participants’ physiological and movement data, including electrodermal activity, heart rate variability, and acceleration, were collected via wearable devices while simultaneously surveying their emotions for data labeling. Machine learning algorithms were used to classify emotions by integrating features derived from sensor data and anomaly detection outcomes. The results demonstrate that integrating data from multiple sensors significantly improved the accuracy of emotion classification compared to relying on single-sensor data alone. The performance was further enhanced by incorporating anomaly features into the analysis. These findings advance the understanding of pedestrians’ subjective emotional experiences and their momentary feelings within pedestrian environments through the continuous application of wearable sensors. This study provides valuable insights into improving walkability by identifying environmental factors and their spatiotemporal characteristics that contribute to pleasant or unpleasant emotional responses in pedestrian environments.