Use of Artificial Neural Network Versus Logistic Regression to Predict Post-Traumatic Mental Disorders

Nowadays, the artificial neural networks have received much attention in predicting the effects of multiple variables and complex relationships in aparticular variables. In this study, we have focused on the use of artificial neural network versus logistic regression to predict post-traumatic mental disorders.
In a prospective cohort study, we covered 100 trauma patients admitted to the trauma center of Shahid Beheshti Hospital of Kashan during a six month period. The patients were then randomly divided into two training (n=50) andexperimental(n=50) groups. 14 variablesincluding age, sex, occupation, education level, marital status, socioeconomic status, history ofmental illnessin theimmediate family, history of being hospitalized in neurosurgeryunit, historyof trauma,history ofunderlying disease, history of psychologicaldrug use, history of anesthesia, history of alcohol use, and history of substance abuse were totally investigated. 300artificial neural networksandlogistic regressions were studied in the first group and then the predicted values were compared in the second group using the two models. The ROC curve and classification accuracy tool were applied to estimate the predictive power of mental disorder.
Findings: The results showed that the accurate index for predicting the disorder was90.65% for the neural network model, while it was 75.96% for the logistic regression model.
Discussion & The artificial neural network models appeared to bemore powerful in predictingmental disorder versus the logistic regression model.