Implementasi Model Transfer Learning EfficientNet untuk Pendeteksian Bahasa Isyarat Indonesia (BISINDO) pada Perangkat Android

Fadhillah, Irnanda Rizka (2024) Implementasi Model Transfer Learning EfficientNet untuk Pendeteksian Bahasa Isyarat Indonesia (BISINDO) pada Perangkat Android. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 20081010171-cover.pdf Download (5MB)
	Text (Bab 1) 20081010171-bab1.pdf Download (242kB)
	Text (Bab 2) 20081010171-bab2.pdf Restricted to Repository staff only until 23 July 2026. Download (630kB)
	Text (Bab 3) 20081010171-bab3.pdf Restricted to Repository staff only until 23 July 2026. Download (796kB)
	Text (Bab 4) 20081010171-bab4.pdf Restricted to Repository staff only until 23 July 2026. Download (2MB)
	Text (Bab 5) 20081010171-bab5.pdf Download (233kB)
	Text (Daftar Pustaka) 20081010171-daftarpustaka.pdf Download (234kB)

Abstract

This research aims to develop an Android application capable of detecting Indonesian Sign Language (BISINDO) using the EfficientNet model through transfer learning techniques. BISINDO, advocated by the Movement for the Welfare of the Deaf Indonesia (GERKATIN), is an important representation of Indonesian Deaf culture. EfficientNet was chosen for its superior balance between accuracy and computational efficiency, which is highly relevant for mobile devices. By leveraging knowledge from large datasets such as ImageNet, it is hoped that this model can enhance performance without straining mobile device resources. By applying transfer learning techniques using EfficientNet and implementing it in this Android application, this research aims to achieve effective and accurate matching of BISINDO hand movements. The model is trained using an 80-20 ratio and three different epoch counts: 20, 30, and 40. The research results indicate that the EfficientNet model can achieve high accuracy in BISINDO recognition. At 20 epochs, the model demonstrates a training accuracy of 99.24% and a validation accuracy of 98.00%, with consistent and stable metrics showing no signs of overfitting. This indicates that the model can effectively learn from the training data and generalize well on unseen data. This research also compares its results with previous studies using TensorFlow Lite EfficientNet, showing that model configuration and fine-tuning significantly impact accuracy outcomes.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Al Haromainy, Muhammad Muharrom	NIDN0701069503	muhammad.muharrom.if@upnjatim.ac.id
Thesis advisor	Maulana, Hendra	NIDN1423128301	hendra.maulana.if@upnjatim.ac.id

Subjects:

T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Mr Irnanda Rizka Fadhillah

Date Deposited:

23 Jul 2024 03:46

Last Modified:

23 Jul 2024 03:46

URI:

https://repository.upnjatim.ac.id/id/eprint/27208