An alternative evaluation of post traumatic stress disorder with machine learning methods

Abstract

In the world we live in, people from different professions are at increased risk for depressive symptoms and posttraumatic stress disorder (PTSD) due to hard working or extreme environmental conditions. Accurate diagnosis and determining the causes are very important to solve these kinds of psychological problems. Machine learning (ML) techniques are gaining popularity in neuroscience due to their high diagnostic capability and effective classification ability. In this paper, alternative hybrid systems which allowed us to develop automatic classifiers for finding the Posttraumatic stress disorder (PTSD) patients are proposed and compared. With the proposed system, not only the PTSD individuals are classified by ML techniques such as sequential minimal optimization (SMO), multilayer perceptron (MLP), Naïve Bayes (NB) but also the important indications of patients' trauma are determined by three popular feature selection methods such as chi-square, principal component analysis (PCA) and correlation based-feature selection (CFS). The effectiveness of the proposed system is examined on a real world dataset. Due to obtained results we can estimate the individuals as PTSD or NONPTSD patients with 74-79% accuracy range, further to that instead of 39 features 7 features are remarked as the most critical symptoms for PTSD.