TELKOMNIKA Telecommunication, Computing, Electronics and Control
Improvement on KNN using genetic algorithm and combined feature extraction to identify COVID-19 sufferers based on CT scan image

TELKOMNIKA Telecommunication, Computing, Electronics and Control
Improvement on KNN using genetic algorithm and combined feature extraction to identify COVID-19 sufferers based on CT scan image

Genetic algorithm
Haralick
Histogram
k-nearest neighbour
Local binary pattern

Coronavirus disease 2019 (COVID-19) has spread throughout the world. The
detection of this disease is usually carried out using the reverse transcriptase
polymerase chain reaction (RT-PCR) swab test. However, limited resources
became an obstacle to carrying out the massive test. To solve this problem,
computerized tomography (CT) scan images are used as one of the solutions
to detect the sufferer. This technique has been used by researchers but mostly
using classifiers that required high resources, such as convolutional neural
network (CNN). In this study, we proposed a way to classify the CT scan
images by using the more efficient classifier, k-nearest neighbors (KNN), for
images that are processed using a combination of these feature extraction
methods, Haralick, histogram, and local binary pattern (LBP). Genetic
algorithm is also used for feature selection. The results showed that the
proposed method was able to improve KNN performance, with the best
accuracy of 93.30% for the combination of Haralick and local binary pattern
feature extraction, and the best area under the curve (AUC) for the
combination of Haralick, histogram, and local binary pattern with a value of
0.948. The best accuracy of our models also outperforms CNN by a 4.3%
margin.

Radityo Adi Nugroho, Arie Sapta Nugraha, Aylwin Al Rasyid, Fenny Winda Rahayu

http://journal.uad.ac.id/index.php/TELKOMNIKA

Aug 5, 2021

peri irawan

pdf

english

text