Knowledge Extraction from Vector Machine Support in the Context of Depression in Children and Adolescents

Abstract

Depression is, worldwide, the main cause of diseases and disabilities during the adolescence. This disorder ails over 300 million people, and can interfere with an individual’s professional performance and education. Therefore, it is essential to conduct research that contributes in the correct diagnosis and treatment of depression, especially on children and adolescents. The Support Vector Machines (SVM) classifier has shown great performance 3 and generalization capabilities when compared to other classifiers, in the context of depression diagnosis. The objective of this study is to explore the depression disorder on children and adolescents, using this classifier. Since the SVM is a black box method, to better understand the model generated we employed the SHAP approach to help explain the model’s output based on feature importance. The final model obtained F-measure results above 87% during training and 82% in its testing. We concluded that the predictive model had satisfactory results and, using the SHAP framework, we explored how the features influenced the results.