Predicting Stress using a Novel Scoring Card Method with Psychological Question
In recent years, machine learning-based predictive methods have seen an increased interest for analyzing quantitative behavioral data in experimental psychology. Although yielding an acceptable accuracy, existing typical prediction methods suffered from pursuit of interpretation ability for further understanding psychological stress questionnaire. This study proposes a novel methodological approach to estimating the propensity scores of 90 composition questions in order to design the prediction method, SCMSTR, based on a scoring card method (SCM) from our dataset which is composed 87 adults without stressful and 82 adults with stressful who responses to the stress scale of 90 questions.SCM calculates the propensities of 90 individual questions to be stressful using statistic discrimination between stressful and not stressful questionnaires of a training data set. Consequently, the propensities of questions are further optimized by maximizing prediction accuracy using an intelligent genetic algorithm. The results show that the simple method SCMSTR with interpretable propensities of questions has promising performance, compared with existing typical prediction methods with blackboxes of outcomes. For predicting stress questionnaires using SCMSTR, has a training accuracy of 100% and a test accuracy of 100%. The propensity score of bottom-rank 2 is “Easily feel exhausted.” which our previous study PreStress had selected it to be an important feature for predicting psychological stress.
Keywords - Stress Questionnaire, Psychologicalstress, Intelligent Genetic Algorithm, Prediction, SCM.