Indonesian Journal of Electrical Engineering and Computer Science Vol. 17, No. 1, January 2020, pp. 524~533 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v17.i1.pp524-533  524 Journal homepage: http://ijeecs.iaescore.com A Multi-layer perceptron based intelligent thyroid disease prediction system Arvind Selwal, Ifrah Raoof Department of Computer Science and Information Technology, Central University of Jammu, J&K, India Article Info ABSTRACT Article history: Received Aug 18, 2018 Revised Jul 6, 2019 Accepted Jul 20, 2019 A challenging task for the modern research is to accurately diagnose the diseases prior to their treatment. Particularly in rural areas, the instant diagnosis for a life style disease is rarely available; it becomes necessary to use modern computing techniques to design intelligent prediction systems. A machine learning model is used for solving complex and non-separable prediction problems in different fields like medical diagnosis, decision support systems, biochemical analysis, image processing and financial analysis etc. The accuracy for thyroid diagnosis system may be improved by considering few additional attributes like heredity, age, anti-bodies etc. In this paper, an improved and intelligent thyroid disease prediction system is developed using multilayer perceptron (MLP) machine learning model. The proposed system uses 7 to 11 features of the individuals to classify them in normal, hyperthyroid and hypothyroid classes. The system uses gradient descent backpropogation algorithm for training the machine learning model using dataset of 120 subjects collected from SKIMS Hospital, Jammu and Kashmir. The thyroid prediction system promises excellent overall accuracy of nearly 99.8% for 11 attributes with more number training instances. However, the system results in a lower accuracy of 66.7% using 11 attributes and 70% using 7 attributes with 30 subjects. Keywords: Intelligent systems Machine learning Multi-layer perceptron Pattern classifier Thyroid disease Copyright © 2020 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Arvind Selwal, Central University of Jammu, Samba, Jammu and Kashmir, India-181143. Email: arvind.cuj@gmail.com 1. INTRODUCTION Machine learning is a modern way of computing where knowledge alongwith a technique is used to build a model which imitates the behaviour of human being. Once the macine learning model is trained it will start predicting the class of a given feature set. As shown in the Figure1, a variety of machine learning techniques are available which may be categorised broadly into supervised, unsupervised and reinforcement learning. The typical examples of supervised machine learning algorithms includes Nearest neighbour classification, regression, Support vector machine (SVM), Artificial neural networks Naïve base classifiers and decision trees. An Artificial neural network (ANN) is an information processing paradigm that is motivated by the way biological neural system i.e. brain process the data. The neural network constitutes of countless interconnected information handling components called neurons. The key component of the neural network is a novel structure. Neural systems, with their efficient capability to derive meaningful information from imprecise information, can be utilized to separate and distinguish patterns that are too intricate to be noticed by any computer technique or by human.As ANN is a self learning framework, it shows distinctive classes of learning calculations, for example, supervised learning, unsupervised learning and reinforcement learning. ANNs are widely used in the real - world computation applications. The various areas of application