Article Per Year (5 Year)
p-Index From 2021 - 2026
0.23
P-Index
This Author published in this journals
All Journal Journal of Electrical Engineering and Informatics
Ferdy Fitriansyah Rowi
Universitas Bengkulu
Computer Science & IT Control & Systems Engineering Electrical & Electronics Engineering Engineering

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

Found 1 Documents
Search

 1 
WORD2VEC Versus Random Embedding Initialization for SARS-COV-2 Genome Classification using CNN Ferdy Fitriansyah Rowi; Ernawati Ernawati; Julia Purnama Sari
Journal of Electrical Engineering and Informatics Vol. 4 No. 1 (2026): Journal of Electrical Engineering and Informatics
Publisher : Fakultas Teknik Universitas Negeri Makassar

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59562/jeeni.v4i1.14557

Abstract

This study evaluates whether contextual Word2Vec initialization provides a consistent performance advantage over Random Embedding initialization for SARS-CoV-2 genome classification using a Convolutional Neural Network (CNN). A computational experiment was conducted using 76,856 cleaned genome sequences from NCBI Virus, consisting of SARS-CoV-2 and non-SARS-CoV-2 respiratory viruses. The data were stratified at the genome level into training, validation, and testing subsets with an 80:10:10 ratio before sliding-window segmentation. Sequences were segmented into 1,000-base windows with a 500-base step and tokenized into overlapping 6-mers. Word2Vec provided pretrained 100-dimensional embedding weights, while Random Embedding used randomly initialized weights with the same vocabulary and dimension. Both embedding layers remained trainable and used the same CNN configuration across three runs. At the window level, Word2Vec achieved balanced accuracy of 0.999572 ± 0.000283 and MCC of 0.984809 ± 0.013068, while Random Embedding achieved 0.999628 ± 0.000076 and 0.988734 ± 0.002149, respectively. PR-AUC for the minority non-SARS-CoV-2 class was 0.997643 ± 0.001402 for Word2Vec and 0.997282 ± 0.002533 for Random Embedding. At the genome level, both approaches produced identical decisions across all three runs and correctly classified 6,743 of 6,744 evaluable genomes. The controlled comparison isolates the effect of embedding initialization and shows that Word2Vec initialization does not provide a consistent performance advantage over trainable Random Embedding under the evaluated configuration.
Co-Authors Ernawati Ernawati Julia Purnama Sari