Predicting visual aesthetic preferences in Tehran city universities campuses using machine learning techniques.

Area of Science:

• Environmental Psychology
• Computer Science
• Landscape Architecture

Background:

• Visual aesthetic preferences significantly influence the restorative potential of university landscapes.
• Student well-being and engagement are demonstrably impacted by campus environmental aesthetics.
• Predicting aesthetic preferences is crucial for designing effective student rest areas.

Purpose of the Study:

• To develop and evaluate Ensemble Learning Models for predicting students' aesthetic preferences in university rest spots.
• To compare the predictive accuracy of ensemble models against conventional individual machine learning models.
• To identify key landscape design elements that contribute to aesthetic appeal and mental restoration.

Main Methods:

• Extracted 18 features from images of 100 student rest spots across four Tehran universities.
• Collected aesthetic preference data from 394 university students.
• Employed Support Vector Regression (SVR), Random Forest (RF), and Multilayer Perceptron (MLP) individually and in ensemble combinations (SVR-MLP, SVR-RF-MLP).

Main Results:

• Ensemble Learning Models demonstrated superior accuracy in predicting aesthetic preferences compared to individual models.
• Individual models showed varying performance: SVR (R²=0.824), MLP (R²=0.814), RF (R²=0.761).
• The SVR-MLP ensemble model achieved the highest accuracy (R²=0.828 on the total dataset).

Conclusions:

• Ensemble learning offers a robust framework for predicting and understanding student aesthetic preferences in campus design.
• Design elements like increased greenery, diverse soft landscapes, water features, and color variety enhance aesthetic appeal and restoration.
• Findings provide actionable insights for architects and planners to create more engaging and restorative university environments.