Article Per Year (5 Year)
p-Index From 2020 - 2025
0.61
P-Index
This Author published in this journals
All Journal MEANS (Media Informasi Analisa dan Sistem) Journal of Mandalika Literature Jurnal Ilmiah Manajemen Dan Kewirausahaan
Abdul Tahir
Unknown Affiliation
Humanities Computer Science & IT Economics, Econometrics & Finance Social Sciences

Published : 3 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search
Journal : MEANS (Media Informasi Analisa dan Sistem)

 1 
Performance Analysis of Neural Networks With Backpropagation on Binary and Multi-Class Data Classification Abdul Tahir; Irdam, Irdam; Sirama , Sirama
MEANS (Media Informasi Analisa dan Sistem) Volume 10 Nomor 1
Publisher : LPPM UNIKA Santo Thomas Medan

Show Abstract | Download Original | Original Source | Check in Google Scholar

Abstract

Neural networks represent a widely adopted paradigm within the domain of machine learning, employed for a multitude of classification endeavors, encompassing image recognition and natural language processing. This investigation seeks to elucidate the influence of varying neuron quantities in hidden layers on the efficacy of neural networks in both binary and multi-class classification endeavors. The research utilizes a dataset procured from images depicting characters and digits, which were transformed into binary format via a thresholding methodology. The neural network architectures comprise one and two hidden layers, which are trained employing the backpropagation algorithm in conjunction with the Adam optimizer. The evaluation of the models is conducted through metrics such as accuracy, loss curves, and confusion matrices. Findings reveal that the configuration featuring two hidden layers with 40 sampai 99 neurons achieves the pinnacle accuracy of 99.64 percent alongside optimal loss stability. Furthermore, models incorporating a single hidden layer exhibited commendable accuracy, thereby indicating that a reduced number of neurons can proficiently encapsulate data complexity in less demanding tasks. This research underscores the criticality of selecting suitable neural network configurations contingent upon data complexity and classification objectives, while advocating for further investigation into regularization strategies to enhance performance.
Co-Authors Amri Yanuar Awwal Hajarul Tahir Irdam, Irdam Israkwaty Muhammad Ardhya Bisma Musakirawaty Sirama , Sirama