Article Per Year (5 Year)
p-Index From 2021 - 2026
0.23
P-Index
This Author published in this journals
All Journal International Journal of Advances in Intelligent Informatics
Lyubomir Gotsev
State University of Library Studies and Information Technologies, Sofia, Bulgaria
Computer Science & IT

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

 1 
Cluster analysis and ensemble transfer learning for COVID-19 classification from computed tomography scans Lyubomir Gotsev; Ivan Mitkov; Eugenia Kovatcheva; Boyan Jekov; Roumen Nikolov; Elena Shoikova; Milena Petkova
International Journal of Advances in Intelligent Informatics Vol 8, No 2 (2022): July 2022
Publisher : Universitas Ahmad Dahlan

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26555/ijain.v8i2.817

Abstract

The paper aims to demonstrate the convergence and synergistic application of deep learning methods and techniques to improve approaches and solutions in COVID-19 computed tomography scans classification by addressing the data and protocol challenges found after analyzing related work. The study is set to test and experiment with the proposed strategy: data standardization and normalization to achieve proper contrast and resolution; k-means (clustering) and group shuffle split to avoid data leakage; augmentation and transfer learning to deal with limited sample size and over-fitting. All activities are implemented before applying ensemble learning. VGG-16, Densenet-201, Inception v3 are the pre-trained networks utilized to build base models for the suggested stacking model with a vector of their predictions fed into a meta-learner input. All four classifiers are measured and compared. Various confusion-matrix-based and weighted evaluation metrics are considered: accuracy, recall, precision, f-measure, specificity, and AUC. Critical measurements, such as negative prediction value, false-positive rate, false-negative rate, and false discovery rate, are also presented. The evaluated classifiers achieve high results with AUC between 0.95 to 1. However, the stacked method is the most reliable. The ensemble approach enhanced described strategy having three main advantages: outperforming the base models, reducing data pitfalls, and decreasing generalization error. It can serve as a baseline to increase the performance quality and mitigate the risk of bias in the field.
Co-Authors Boyan Jekov Elena Shoikova Eugenia Kovatcheva Ivan Mitkov Milena Petkova Roumen Nikolov