Two- Convolutional neural network based on the Goldberg scale for the classification of anxiety and depression

Abstract

Anxiety and depression are highly prevalent and frequently comorbid mental disorders, suggesting the existence of shared cognitive and affective patterns. The Goldberg Anxiety and Depression Scale (GADS) is a widely used tool for their detection, supported by adequate validity and reliability indices in the Salvadoran population. However, the need to optimize screening processes justifies the search for methods that reduce the length of questionnaires without compromising their diagnostic capacity. The objective of this study was to evaluate whether the items of one GADS subscale are sufficient to predict the presence of symptoms in the opposite subscale using artificial neural networks. Two models were trained: one that used anxiety items to predict depressive symptoms and another that used depression items to predict anxiety. Both models showed solid performance, with F1-scores between 0.84 and 0.85 and AUC values above 0.82, demonstrating adequate discriminatory capacity. The results indicate that it is possible to significantly reduce the number of questions while maintaining accurate classification, which opens the door to the development of abbreviated versions based on deep learning for rapid mental health screening.