Recently, as interest in addressing climate change and deepening urbanization has grown, innovative AI (Artificial Intelligence)-based solutions have been actively discussed in the landscaping and forest management sectors. However, research focused o...
Recently, as interest in addressing climate change and deepening urbanization has grown, innovative AI (Artificial Intelligence)-based solutions have been actively discussed in the landscaping and forest management sectors. However, research focused on recommending tree species optimized for on-site environmental conditions remains relatively insufficient. Existing tree recommendation methods rely on experiential knowledge and manual processing, failing to adequately reflect complex environmental factors and lacking a cyclical learning function, which limits performance advancement.
Therefore, this study aimed to propose an efficient alternative for precise response to climate change and urbanization by developing a User-Centric and AI-Powered Custom Tree Recommendation System Focused on UI/UX.
First, a high-quality dataset optimized for AI model training was constructed. Public data and literature related to landscaping trees were collected, and raw data was refined into effective data through preprocessing and validation processes. Specifically, a database including over 57 environmental variables, such such as temperature (mean/extreme), precipitation, soil characteristics (pH, Organic Matter, CEC), and topography (elevation, slope, aspect), was established to enable the model to precisely learn complex environmental factors.
Second, a LightGBM (Light Gradient Boosting Machine)-based Decision Tree ensemble model was developed. The final performance was measured using an independent validation set ($\text{6,110}$ records), completely separated from the training set ($\text{18,327}$ records) using the Hold-out data partitioning method (Total Dataset: $\text{24,437}$ records). Accuracy was measured as the ratio of correct predictions to total predictions, and Precision was calculated as the probability of recommending suitable trees while excluding unnecessary ones. Through this rigorous validation, the model achieved a performance of 98% Accuracy and 98% Precision. This significantly exceeds the initial goals of 95% Accuracy and 85% Precision, proving the AI model's ability to provide accurate tree recommendations by comprehensively considering diverse environmental variables. The model demonstrated prediction capabilities that incorporate tree growth suitability, carbon storage, and indirect external factors.
Third, the system architecture was designed using a modern technology stack, including Python, FastAPI (Fast Application Programming Interface), Docker, and AWS (Amazon Web Services), and a User-Friendly UI/UX (User Interface/User Experience) was implemented. The tree recommendation system is structured as a Closed-loop process, ranging from input to planting registration and Cyclical Learning. This design provides a direct screen flow and simplified interaction, enabling not only landscaping experts but also general users to easily utilize accurate recommendation results without complex analysis.
In summary, this research presents a recommendation system that integrates a high-quality data infrastructure, an explainable learning and inference framework, and a Cyclical Learning-based feedback structure. This contributes to the standardization of data-driven decision-making in the landscaping and forestry sectors and proves its potential for application expansion into urban green space management, climate-resilient landscape design, and carbon-neutral strategies. Future tasks include continuous enhancement of generalization performance and field impact through model advancement, real-time data integration, deep learning and multi-modal analysis, and user experience improvement.