A neuro-fuzzy approach for the diagnosis of depression

Depression is considered to be a chronic mood disorder. This paper attempts to mathematically model how psychiatrists clinically perceive the symptoms and then diagnose depression states. According to Diagnostic and Statistical Manual (DSM)-IV-TR, fourteen symptoms of adult depression have been considered. The load of each symptom and the corresponding severity of depression are measured by the psychiatrists (i.e. the domain experts). Using the Principal Component Analysis (PCA) out of fourteen symptoms (as features) seven has been extracted as latent factors. Using these features as inputs, a hybrid system consisting of Mamdani's Fuzzy logic controller (FLC) on a Feed Forward Multilayer Neural Net (FFMNN) has been developed. The output of the hybrid system was tuned by a back propagation (BPNN) algorithm. Finally, the model is validated using 302 real-world adult depression cases and 50 controls (i.e. normal population). The study concludes that the hybrid controller can diagnose and grade depression with an average accuracy of 95.50%. Finally, it is compared with the accuracies obtained by other techniques.