Implementación de técnicas de Machine learning para el diagnóstico automático de trastornos mentales

Abstract

Generating a clinical diagnosis of a mental disorder is a complex process due to the variety of biological factors that affect this type of condition, so it is necessary that a professional performs a deep evaluation in order to identify and determine the type of disorder that affects the patient. This paper proposes the implementation and comparison of five machine learning algorithms (ML) to generate automatic diagnoses of mental disorders, through the set of symptoms present in a patient. The algorithms selected for comparison are: Vector Support Machine (SVM), Logistic Regression, Random Forest, Bayesian Networks, Nearest K-neighbors (k-NN). Also, a web portal was developed, which allowed the acquisition of essential data for the training of the models. The evaluation was executed through performance, accuracy, recall, score and accuracy metrics. The general results show that the Logistic Regression algorithm obtained a better performance showing 55% and 71% accuracy. The SVM model on the other hand showed a low performance reaching only 25% accuracy.