Toward a Machine Learning Predictive-Oriented Approach to Complement Explanatory Modeling. An Application for Evaluating Psychopathological Traits Based on Affective Neurosciences and Phenomenology

This paper presents a procedure that aims to combine explanatory and predictive modeling for the construction of new psychometric questionnaires based on psychological and neuroscientific theoretical grounding. It presents the methodology and the results of a procedure for items selection that considers both the explanatory power of the theory and the predictive power of modern computational techniques, namely exploratory data analysis for investigating the dimensional structure and artificial neural networks (ANNs) for predicting the psychopathological diagnosis of clinical subjects. Such blending allows deriving theoretical insights on the characteristics of the items selected and their conformity with the theoretical framework of reference. At the same time, it permits the selection of those items that have the most relevance in terms of prediction by therefore considering the relationship of the items with the actual psychopathological diagnosis. Such approach helps to construct a diagnostic tool that both conforms with the theory and with the individual characteristics of the population at hand, by providing insights on the power of the scale in precisely identifying out-of-sample pathological subjects. The proposed procedure is based on a sequence of steps that allows the construction of an ANN capable of predicting the diagnosis of a group of subjects based on their item responses to a questionnaire and subsequently automatically selects the most predictive items by preserving the factorial structure of the scale. Results show that the machine learning procedure selected a set of items that drastically improved the prediction accuracy of the model (167 items reached a prediction accuracy of 88.5%, that is 25.6% of incorrectly classified), compared to the predictions obtained using all the original items (260 items with a prediction accuracy of 74.4%). At the same time, it reduced the redundancy of the items and eliminated those with less consistency.