Psychological mental health questionnaire refinement through dimensionality reduction using machine learning and XAI techniques

Mental disorders are a worldwide problem. In the USA alone, approximately one billion dollars are spent annually in order to combat the negative effects that mental disorders cause. One of these is that it can lead to suicide, which is the second leading cause of death in young adults and adolescents. Nevertheless, the early detection of mental disorders considerably increases the probability of recovery, so the use of self-assessment psychological questionnaires aimed at detecting mental disorders early is crucial. Despite this, and due to the large number of questionnaires that exist for the multiple symptoms that they assess, the combined use of these different questionnaires creates a considerable number of total questions, which reduces the quality of the responses. In this work, the use of a ML model is proposed, so that, together with the application of XAI techniques (specifically ALE, LIME, SHAP and FI), the weight and relevance of each item of the questionnaires are determined, in order to eliminate those that may lead to an erroneous classification. The aforementioned process is applied iteratively and cyclically until no more questions are found to delete. Said ML model was trained through the responses of 160 undergraduate students to the combined use of different questionnaires commonly used for screening depression. As a result, the refined questionnaire and different graphs will be obtained where the progress and modifications made in each iteration will be appreciated.