Prediction of mental disorders after Mild Traumatic Brain Injury: principle component Approach

Introduction: In Processes Modeling, when there is relatively a high correlation between covariates, multicollinearity is created, and it leads to reduction in model's efficiency. In this study, by using principle component analysis, modification of the effect of multicolinearity in Artificial Neural Network (ANN) and Logistic Regression (LR) has been studied. Also, the effect of multicolinearity on the accuracy of prediction of mental disorders after trauma in patients with Mild Traumatic Brain Injury has been investigated. Methods: In a prospective cohort Study, first, during 6 months period, 100 patients with Mild Traumatic Brain Injury have been selected. Then, by using Primary Covariates and Principle Component Analysis, Logistic Regression and ANN models have been conducted and based on these models prediction have been done. (Receiver Operating Characteristic) ROC curve and Accuracy Rate have been used to compare the strength of model’s prediction. Results: The results revealed that Accuracy Rate for ANN before and after applying principle component analysis are 84.22 and 91.23% respectively, and for Logistic Regression models are 72.33% and 74.89% respectively. Conclusion: The study showed that the Accuracy Rate was higher for models based on Principle Component Analysis including primary covariates; hence, when multicolinearity exists, models that use the principle component for prediction of mental disorders are more effective compare to other methods. Also, ANN Models are more effective than Regression models.