Urban trees offer essential ecosystem benefits but can pose safety hazards if they fail structurally. The traditional Preliminary Tree Assessment (PTA) serves as a rapid, non-invasive screening tool; however, its outcomes are often subjective and reliant on the assessor’s expertise. This study introduces a species-specific, machine learning–based protocol employing meta-decision tree classifiers to enhance PTA-level evaluation of tree failure risk in urban environments. Field data from 1,193 trees of four common campus species—Delonix regia, Pterocarpus indicus, Terminalia ivorensis, and Lagerstroemia speciosa—at Thammasat University, Thailand, were collected using standardized visual indicators weighted by expert arborists. Separate decision trees were trained per species and integrated into a meta-classifier to categorize risk as unlikely, likely, or possible failure. Performance metrics, including confusion matrices and Cohen’s kappa, demonstrated an overall accuracy of 79.03% (κ=0.580), with better reliability for unlikely and likely failure categories, although possible failures remained underdetected. Results highlight significant variation in failure patterns and defects among species, underscoring the value of species-specific models. While this framework enhances objectivity and efficiency at the PTA stage, it does not replace advanced diagnostics. The approach facilitates data-driven urban tree management prioritization, acknowledging the limitations of AI screening tools.
Srivanit et al. (Tue,) studied this question.