This study addresses a critical gap by investigating the subjective perception of environmental barriers among youth with physical disabilities, moving beyond objective cataloging. It further examines the extent to which these perception profiles can be predicted using a hybrid machine learning approach. Drawing on data from 100 youth with physical disabilities in Garzon – Huila, Colombia, the study first employed unsupervised clustering (K-Means) to identify three distinct barrier perception profiles: Low, Moderate, and High. Subsequently, seven supervised learning models were trained to predict profile membership. The Gradient Boosting Classifier (GBC) emerged as the top-performing model. Its predictive power was significantly enhanced by isolating a “predictive triad” of variables aligned with key domains of the International Classification of Functioning, Disability and Health (ICF): self-reported difficulties with Mobility, Cognition, and Communication. Using this optimized feature set, the model achieved an overall accuracy of 92%, with exceptionally high F1-Scores and ROC-AUC values (0.98–1.00). The study demonstrates that barrier perception is a predictable and multidimensional construct, extending beyond physical limitations to encompass navigational and social interaction challenges. These findings provide an empirical framework for developing personalized interventions, offering a data-driven alternative to “one-size-fits-all” policies.