Using Machine Learning to Optimize the Visual Perceptual Environment of Inter-house Leisure Spaces in Cold Winter Environments

In the winter environment of cold regions, the visual perception of inter-house leisure space in settlements has an important impact on the physical and mental health of residents. The study used an eye-tracking device combined with a semantic differential (SD) questionnaire to screen the visual attention elements and attributes of winter residents in cold regions, including the spatial aspect ratio (D/H), residential elevation saturation (RES), and the percentage of lawn in the field of view (POL). Orthogonal experiments were established in an immersive virtual environment to reveal the influence mechanism of visual perceptual environmental factors of cold regions settlements on residents' leisure space evaluation in winter environment. The study trained and compared four visual perception machine learning agent models, combining genetic algorithms (GA) with k-nearest neighbor algorithms (KNN), resulting in optimized threshold ranges of D/H: 2.22-2.54, RES: 68.47-82.34, and POL: 10%-14% in winter environments.