Deep Learning–Based Detection of Anxiety and Perfectionism Patterns in High-Ability Youth

The objective of this study was to identify and classify latent patterns of anxiety and perfectionism among high-ability adolescents using a multimodal deep learning framework. This quantitative, cross-sectional study was conducted on a sample of high-ability youth aged ۱۲ to ۱۸ years in Germany. Participants were recruited from secondary schools and gifted education programs and completed standardized self-report measures assessing anxiety and multidimensional perfectionism through a secure digital platform. In addition to numerical questionnaire data, open-ended textual responses related to academic experiences and self-expectations were collected, along with behavioral interaction indicators such as response times. Data were analyzed using a multimodal deep learning architecture integrating feedforward neural networks for numerical features and transformer-based models for textual data. Feature fusion was performed in a shared latent space, and supervised learning was applied to classify anxiety–perfectionism profiles. Model performance was evaluated using accuracy, precision, recall, F۱-score, and area under the receiver operating characteristic curve. Inferential analyses indicated that the multimodal deep learning model significantly outperformed single-modality models in detecting anxiety and perfectionism patterns. Latent profile analysis based on learned representations revealed three distinct psychological profiles: low-anxiety adaptive perfectionism, moderate mixed perfectionism, and high-anxiety maladaptive perfectionism. The high-anxiety maladaptive profile constituted the largest subgroup, and linguistic and behavioral features contributed significantly to classification accuracy beyond self-report measures alone. The findings demonstrate that multimodal deep learning approaches can effectively uncover nuanced and clinically meaningful anxiety–perfectionism profiles in high-ability youth, offering a robust foundation for early identification and targeted psychological support. The objective of this study was to identify and classify latent patterns of anxiety and perfectionism among high-ability adolescents using a multimodal deep learning framework. This quantitative, cross-sectional study was conducted on a sample of high-ability youth aged ۱۲ to ۱۸ years in Germany. Participants were recruited from secondary schools and gifted education programs and completed standardized self-report measures assessing anxiety and multidimensional perfectionism through a secure digital platform. In addition to numerical questionnaire data, open-ended textual responses related to academic experiences and self-expectations were collected, along with behavioral interaction indicators such as response times. Data were analyzed using a multimodal deep learning architecture integrating feedforward neural networks for numerical features and transformer-based models for textual data. Feature fusion was performed in a shared latent space, and supervised learning was applied to classify anxiety–perfectionism profiles. Model performance was evaluated using accuracy, precision, recall, F۱-score, and area under the receiver operating characteristic curve. Inferential analyses indicated that the multimodal deep learning model significantly outperformed single-modality models in detecting anxiety and perfectionism patterns. Latent profile analysis based on learned representations revealed three distinct psychological profiles: low-anxiety adaptive perfectionism, moderate mixed perfectionism, and high-anxiety maladaptive perfectionism. The high-anxiety maladaptive profile constituted the largest subgroup, and linguistic and behavioral features contributed significantly to classification accuracy beyond self-report measures alone. The findings demonstrate that multimodal deep learning approaches can effectively uncover nuanced and clinically meaningful anxiety–perfectionism profiles in high-ability youth, offering a robust foundation for early identification and targeted psychological support.

Keywords:

High-ability youth , Anxiety , Perfectionism , Deep learning , Multimodal analysis