Application of random forest algorithm in the study of microscopic features and visitor experience evaluation in gardens

Abstract In the development of modern urban gardens, delving into visitor experience perception to address homogenization issues is a crucial means to enhance garden space characteristics and meet diverse demands. This study selected nine representative modern gardens in the Taihu Lake region as samples. Utilizing machine learning techniques combined with image semantic segmentation and principal component analysis, it systematically assessed various microscopic features of gardens and explored their relationship with actual visitor experience ratings. The aim is to reveal emotional resonance and aesthetic preferences during the touring process. The results indicate: (1) Comfort holds a primary position in influencing visitor touring experiences; (2) Microscopic features such as water features, tall vegetation, architecture, and ornametal elements significantly contribute to visitor comfort; (3) The random forest model unveils complex interactions between microscopic features and the overall visitor experience, indicating limitations in traditional linear model evaluations for such relationships. Through analyzing the relationship between microscopic garden features and visitor experiences in the selected cases, this study highlights the importance of adopting non-linear models to optimize garden landscape design and enhance the visitor touring experience.