چكيده به لاتين
ABSTRACT
Diabetes is the most prevalent endocrine disease in the world and is the fourth or fifth leading cause of death in most counties. Timely diagnosis of diabetes plays a major role in reducing its complications, and the patient can live a full life by taking medications and changing lifestyle. In the present study, data mining predictive models were used to identify individuals with diabetes. The objective of this study was to increase the accuracy of identification of diabetes using a Hybrid classification approach based on assigning weights to the classifiers. The proposed method includes the following stages: feature extraction, division of the problem space, feature selection in each subspace, and assigning weights to the predictive algorithms in each subspace. In order to enrich the feature space, a limited number of features was extracted using the MVU algorithm to be added to the primary features of the diabetes dataset. Then, kmeans clustering algorithm was applied to divide the primary space of the problem into subspaces. A feature selection was then performed using the Hilbert-Schmidt Independence Criterion in each subspace to select the features with the highest association with the label feature class.As the last stage, weights were assigned to the predictive algorithms in the subspaces.
In this study, the PIMA Indian Diabetes dataset from the UCI machine learning database was employed. Experimental results on the diabetes test data show that the proposed method is more accurate than existing methods.
KEYWORDS
Feature extraction,Clustering, Feature selection, Weighting to classifiers, predictive algorithms.
