Predicting Diabetes Disease Applying Neural Network Using Data Visualisation Technique

Abstract

The purpose of the paper is to carry out the medical diagnosis system in detection of heart disease.  Such diagnosis system is not much implemented due to the scarcity of doctors and ignorance of the patients. The visualization techniques for multi-dimensional heart data set, such as parallel coordinates are plotted for large number of records with number of dimensions. But, when the dimensions are more; it is very difficult to understand, hence, Visualisation for Dimension Reduction Technique (VDRT) is made use to understand the similarity of pattern more effectively.  Parallel Coordinate is the most important techniques in Visual Pattern Recognition with Dimension Reduction technique using spline is computed. However, it could visually show the cluster of patients who are having heart disease, to make clear to uneducated peoples. Hence, the Rule-Based model is computed along with the unsupervised models such as, Principle Component Analysis (PCA), Support Vector Machines (SVM) and Kohenen Self-Organizing Method (KSOM), which is employed to compute the heart disease chances of occurrence to a patient more accurately. The comparative results are tabulated and it reveals that the KSOM model yields better performance. The popular heart dataset having 14 dimensions and 271 records are used for testing information visualisation using the tool namely, Orange. In this paper, an effort is made to remove the number of dimensions by reducing the 7 unimportant dimensions. Both the Orange tool and the datasets are available as open source for public domain, which is considered as one of the popular benchmark to have a better perspective of information visualisation.