Active Learning dengan Strategi Seleksi Berbasis Probabilitas Prediksi pada Klasifikasi Tumbuhan Kebun Raya Balikpapan Menggunakan MobileViT V2 - Submit Jurnal

As-Sddiq, Fhiqi Maulana (2026) Active Learning dengan Strategi Seleksi Berbasis Probabilitas Prediksi pada Klasifikasi Tumbuhan Kebun Raya Balikpapan Menggunakan MobileViT V2 - Submit Jurnal. Bachelor thesis, Institut Teknologi Kalimantan.

[img] Text
11221003_cover.pdf
Download (538kB)
[img] Text
11221003_statement_of_authenticity.pdf
Download (207kB)
[img] Text
11221003_publishing_agreement.pdf
Download (246kB)
[img] Text
11221003_approval_sheet.pdf
Download (359kB)
[img] Text
11221003_preface.pdf
Download (245kB)
[img] Text
11221003_abstract_id.pdf
Download (275kB)
[img] Text
11221003_abstract_en.pdf
Restricted to Repository staff only until 3 January 2028.
Download (303kB) | Request a copy
[img] Text
11221003_table_of_content.pdf
Restricted to Repository staff only until 3 January 2028.
Download (404kB) | Request a copy
[img] Text
11221003_illustrations.pdf
Restricted to Repository staff only until 3 January 2028.
Download (353kB) | Request a copy
[img] Text
11221003_tables.pdf
Restricted to Repository staff only until 3 January 2028.
Download (351kB) | Request a copy
[img] Text
11221003_notations.pdf
Restricted to Repository staff only until 3 January 2028.
Download (301kB) | Request a copy
[img] Text
11221003_chapter_1.pdf
Restricted to Repository staff only until 3 January 2028.
Download (428kB) | Request a copy
[img] Text
11221003_chapter_2.pdf
Restricted to Repository staff only until 3 January 2028.
Download (674kB) | Request a copy
[img] Text
11221003_chapter_3.pdf
Restricted to Repository staff only until 3 January 2028.
Download (609kB) | Request a copy
[img] Text
11221003_chapter_4.pdf
Restricted to Repository staff only until 3 January 2028.
Download (544kB) | Request a copy
[img] Text
11221003_conclusions.pdf
Restricted to Repository staff only until 3 January 2028.
Download (332kB) | Request a copy
[img] Text
11221003_bibliography.pdf
Download (266kB)
[img] Text
11221003_enclosure.pdf
Restricted to Repository staff only until 3 January 2028.
Download (214kB) | Request a copy
[img] Text
11221003_paper.pdf
Restricted to Repository staff only until 3 January 2028.
Download (586kB) | Request a copy
[img] Text
11221003_presentation.pdf
Restricted to Repository staff only until 3 January 2028.
Download (1MB) | Request a copy
[img] Text
11221003_Form. TA-020.pdf
Restricted to Repository staff only until 3 January 2028.
Download (603kB) | Request a copy

Abstract

Identifikasi spesies tumbuhan di Kebun Raya Balikpapan (KRB) secara manual terkendala terbatasnya pakar taksonomi, lambatnya validasi, dan mahalnya biaya operasional. Otomatisasi dengan deep learning menawarkan solusi, namun terhambat mahalnya biaya pelabelan data skala besar. Penelitian ini bertujuan mengembangkan sistem klasifikasi tumbuhan yang efisien menggunakan arsitektur model ringan MobileViT v2 dan metode active learning. Tiga strategi seleksi (prediction probability-based selection, hybrid sampling, dan random sampling) dievaluasi menggunakan dataset BBG52 (52 spesies) dengan variasi AL batch size 16, 32, 64, dan 128 sampel per iterasi. Evaluasi performa model diukur menggunakan metrik F1-Score. Hasil eksperimen menunjukkan bahwa strategi seleksi berbasis informativitas data jauh lebih efisien menekan kebutuhan data berlabel dibandingkan random sampling. Pada AL batch size kecil hingga menengah (16 dan 32), strategi hybrid sampling menghasilkan performa terbaik dengan capaian F1-Score tertinggi masing-masing sebesar 0,820 (pada 2.000 data berlabel) dan 0,818 (pada 2.144 data berlabel). Sementara itu, prediction probability-based selection unggul pada batch size 64 dengan F1-Score 0,815 menggunakan 1.888 data berlabel. Pembesaran ukuran batch hingga 128 menurunkan keunggulan strategi seleksi informatif akibat akumulasi data yang terlalu besar dalam satu putaran. Kesimpulannya, integrasi active learning dan MobileViT v2 terbukti menghasilkan model klasifikasi dengan performa tinggi menggunakan jumlah data berlabel yang lebih sedikit, sehingga menjadi solusi praktis dan hemat biaya untuk digitalisasi koleksi tumbuhan di Kebun Raya Balikpapan.

Item Type: Thesis (Bachelor)
Subjects: T Technology > T Technology (General)
Divisions: Jurusan Matematika dan Teknologi Informasi > Informatika
Depositing User: Fhiqi Maulana As-Sddiq
Date Deposited: 13 Jul 2026 06:37
Last Modified: 13 Jul 2026 06:37
URI: http://repository.itk.ac.id/id/eprint/26753

Actions (login required)

View Item View Item