Adaptive Neuro Fuzzy Inference System for Diagnosis of Stimulant Use Disorders

Abstract ―​

The main challenge within the health sector presently is the need for a quick, lower cost and robust disease diagnosis techniques. Adaptive-neuro-fuzzy inference system (ANFIS) models are recognized as efficient models among neuro-fuzzy techniques and additionally amongst different machine learning systems for its learning capability and unambiguous information representation. This study designed an ANFIS model for the detection of stimulant use disorders using hybrid learning techniques to improve the diagnosis based on the conventional model. The system represents stimulant use disorders with 15 symptoms and one output. The study used first-order Sugeno fuzzy system to provide the rule base of the system. The Gaussian membership function was utilized for input and linear membership function was considered for output parameters. The hybrid learning method was employed, comprising of least squares method, and the gradient descent method. The model was trained and validated using clinical data from Specialist Hospital, Yola Nigeria. The performance of model was evaluated in terms of the prediction errors. The results revealed that the system gave an accuracy of 95.6% in detecting severity of stimulant use disorders, indicating that the technique can be very useful in psychological problem detection with minimal errors.