Volume 18, No. 6, 2021
Classification For Risk Factor Identification And Disease Diagnosis Using Kernelized Normal Discriminant Feature Selection And Borda Count Bootstrap Aggregating Classification
P.S RENJENI , B. MUKUNTHAN , B. SENTHILKUMARAN , L. JAYA SINGH DHAS
Abstract
In order to examine vast amounts of patient data, automatic illness detection is critical in health care administration. The goal of early disease identification and treatment is critical in preventing the patient's death. The development of disease diagnosis has been aided by a number of researchers. However, it increases the danger of misdiagnosing a patient's health condition. A Kernelized Normal Discriminant Feature Selection based Borda count bootstrap aggregating Classification (KNDFS-BCBAC) technique is introduced to increase illness diagnosing accuracy by detecting the patient's health state and crucial factor analysis with higher accuracy and less time. To reduce the complexity of disease diagnosis, radial basis kernelized normal discriminant analysis is initially utilised to locate the relevant feature. By creating the weak learner as a bivariate correlated regression tree, Borda count bootstrap aggregating Classifier is used to categorise the patient data as abnormal or normal after picking the appropriate features. The diseased data is then used as a training sample for analysing the crucial factor, and the patient data level is classified as either initial or critical depending on the feature value threshold range. The Borda count voting technique is used to aggregate the weak learner results into strong results. Disease diagnosis and crucial factor analysis of patient data are performed with greater accuracy and less time complexity in this manner. With respect to a variety of patient data, an experimental evaluation is conducted with a tumour dataset on parameters such as illness diagnosis accuracy, false alarm rate, and time complexity. The findings reveal that the KNDFS-BCBAC strategy achieves higher illness diagnostic accuracy with less complexity and a lower false-alarm rate than current methods.
Pages: 7573-7593
Keywords: Disease diagnosis, feature selection, kernelized normal discriminant analysis, bootstrap aggregating Classification, bivariate correlated regression tree, Borda count voting scheme.