Article Per Year (5 Year)
p-Index From 2021 - 2026
2.48
P-Index
This Author published in this journals
All Journal Informatika Mulawarman: Jurnal Ilmiah Ilmu Komputer RABIT: Jurnal Teknologi dan Sistem Informasi Univrab JIKO (Jurnal Informatika dan Komputer) JOURNAL OF APPLIED INFORMATICS AND COMPUTING KOMIK (Konferensi Nasional Teknologi Informasi dan Komputer) JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika) Vocatech : Vocational Education and Technology Journal TECHSI - Jurnal Teknik Informatika Lok Seva: Journal of Contemporary Community Service Journal Of Artificial Intelligence And Software Engineering Jurnal Pengabdian Sosial Journal of Deep Learning, Computer Vision and Digital Image Processing Journal of Advanced Computer Knowledge and Algorithms Seuraya Jurnal Pengabdian Kepada Masyarakat Proceedings of International Conference on Multidisciplinary Engineering (ICOMDEN) Jurnal Pengabdian Masyarakat Teknologi Informasi (PERANTI)
Hamdhana, Defry
Universitas Malikussaleh
Religion Humanities Chemical Engineering, Chemistry & Bioengineering Chemistry Civil Engineering, Building, Construction & Architecture Computer Science & IT Control & Systems Engineering Decision Sciences, Operations Research & Management Economics, Econometrics & Finance Education Electrical & Electronics Engineering Engineering Industrial & Manufacturing Engineering Law, Crime, Criminology & Criminal Justice Mechanical Engineering Public Health Social Sciences Other

Published : 27 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 27 Documents
Search

 1   2   3 
ANALISIS PERBANDINGAN KINERJA ALGORITMA AGGLOMERATIVE HIERARCHICAL CLUSTERING DAN K-MEDOIDS UNTUK KLASTERISASI JENIS PENYAKIT PASIEN RAWAT INAP: COMPARATIVE ANALYSIS OF THE PERFORMANCE OF AGGLOMERATIVE HIERARCHICAL CLUSTERING AND K-MEDOIDS ALGORITHM FOR CLUSTERING DISEASE TYPES OF INPATIENTS Lailatul Husna; Defry Hamdhana; Munirul Ula
Rabit : Jurnal Teknologi dan Sistem Informasi Univrab Vol 10 No 2 (2025): Juli
Publisher : LPPM Universitas Abdurrab

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36341/rabit.v10i2.6554

Abstract

Rumah Sakit Arun Lhokseumawe memiliki data jenis penyakit pasien rawat inap yang beragam dan terus bertambah, namun belum dimanfaatkan secara optimal untuk analisis lebih lanjut. Selama ini, data hanya menjadi laporan administratif tanpa dilakukan pengolahan guna memperoleh informasi bermakna, seperti pola dominasi penyakit. Riset ini bertujuan untuk mengelompokkan jenis penyakit pasien rawat inap menggunakan dua metode klasterisasi, yaitu Agglomerative Hierarchical Clustering dan K-Medoids, serta menganalisis perbandingan kinerjanya. Data yang dipergunakan mencakup 84 jenis penyakit yang direkam pada periode Desember 2024 hingga Januari 2025, dengan atribut jumlah pasien laki-laki, pasien perempuan, dan umur pasien. Klasterisasi Agglomerative Hierarchical Clustering dilakukan dengan pendekatan average linkage dan jarak Manhattan Distance, sedangkan K-Medoids menggunakan jarak Euclidean Distance. Hasil memperlihatkan bahwa metode Agglomerative Hierarchical Clustering membentuk 3 cluster, yaitu C1 dengan 4 jenis penyakit, C2 menghasilkan 79 jenis penyakit, dan C3 menghasilkan 1 jenis penyakit. Sedangkan metode K-Medoids juga menghasilkan 3 cluster dengan C1 dengan 11 jenis penyakit, C2 menghasilkan 13 jenis penyakit, dan C3 menghasilkan 60 jenis penyakit. Evaluasi hasil dilakukan menggunakan Silhouette Coefficient. Berdasarkan pengujian validitas cluster Agglomerative Hierarchical Clustering menggunakan Silhouette Coefficient, algoritma Agglomerative Hierarchical Clustering memperlihatkan kinerja lebih baik dengan rerata nilai 0,5837. Sedangkan pengujian validitas cluster K-Medoids menggunakan Silhouette Coefficient pada seluruh data, diperoleh nilai rerata sejumlah -0.3558. Nilai ini memperlihatkan bahwa hasil kualitas cluster yang kurang optimal. Perbedaan hasil klasterisasi antara AHC dan K-Medoids terjadi karena kedua algoritma memiliki cara kerja yang berbeda. AHC membentuk cluster secara bertahap dengan menggabungkan data yang paling dekat satu per satu hingga membentuk struktur hierarki menggunakan jarak Manhattan, sedangkan K-Medoids langsung membagi data ke dalam jumlah cluster di awal dengan menggunakan jarak Euclidean. Perbedaan ini memengaruhi jumlah dan susunan anggota pada tiap cluster. Kata Kunci: Klasterisasi, Agglomerative Hierarchical Clustering, K-Medoids, Silhouette Coefficient
Integration of named entity recognition and latent Dirichlet allocation for extracting cyberbullying issues on X Juanda Pratama; Defry Hamdhana; Zara Yunizar
Journal of Deep Learning, Computer Vision, and Digital Image Processing Volume 4 Issue 2 June 2026
Publisher : CV. Sakura Digital Nusantara

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.61255/decoding.v4i2.1528

Abstract

Purpose – The rapid growth of social media Platform X has increased the risk of cyberbullying, which is difficult to detect due to the unstructured nature of textual data. This study proposes an integration of Named Entity Recognition (NER) and Latent Dirichlet Allocation (LDA) to support the extraction of cyberbullying-related social issues.Methods – A total of 2,000 tweets were processed through preprocessing, spaCy-based entity extraction, TF-IDF weighting, and LDA topic modeling. The latent topics generated by LDA were manually mapped into four predefined categories (Bodyshaming, Racism, Gender, and Neutral) and evaluated against researcher-annotated ground truth labels.Findings – Experimental results achieved an overall accuracy of 80%, with F1-scores of 94% for Racism, 93% for Gender, 70% for Bodyshaming, and 63% for Neutral.Research implications – The proposed framework provides practical support for monitoring cyberbullying patterns and assisting policymakers in understanding online social issues.Originality – The originality of this research lies in the sequential integration of NER as an entity-filtering stage prior to LDA, enabling a more comprehensive analysis of cyberbullying discussions than the isolated application of either method.
Democratizing Climate Intelligence Through Localizez Large Language Models for Education and Governance Zulfikar Aji Kusworo; Widyana Verawaty Siregar; Baharuddin Ismail; Defry Hamdhana
VOCATECH: Vocational Education and Technology Journal Vol 8, No 1 (2026): April
Publisher : Akademi Komunitas Negeri Aceh Barat

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.38038/vocatech.v8i1.325

Abstract

 AbstractClimate change presents complex challenges in Indonesia, where local governments and communities frequently experience information asymmetry and limited access to expert knowledge, particularly in low-resource and low-connectivity regions. This study aims to develop and evaluate a localized, domain-adapted Large Language Model (LLM) that functions as offline-capable climate knowledge infrastructure for education and local governance in Indonesia. The research method employs a design science research methodology comprising four stages: (1) selection of Qwen3-4B as the base model, (2) curation of an Indonesian climate and energy transition corpus containing approximately 12,400 instruction-response pairs (~38 MB) drawn from national climate policy documents, NDC/RPJMN frameworks, renewable energy guidelines, and educational climate science texts, (3) parameter-efficient fine-tuning using QLoRA with LoRA rank r=16, alpha=32, learning rate 2e-4, 3 epochs, per-device batch size 2 with gradient accumulation 4, and 4-bit NF4 quantization, and (4) offline deployment on consumer-grade hardware with task-oriented evaluation against three baseline models (Qwen3-4B-Thinking, Gemma-3-4B, LLaMa-3.1-8B). The results show that the fine-tuned model (Qwen3-4B-REnewbie v1) achieved a 15.4% perplexity reduction on domain-specific test data and an average qualitative score of 9.3/10 across factual accuracy, reasoning structure, and Bahasa Indonesia language compliance, outperforming all baselines (score range 7.0–8.2). The system operates fully offline on consumer-grade hardware with acceptable inference latency. The conclusion drawn from this study is that localized, resource-efficient LLMs can function as practical climate knowledge infrastructure for vocational education and local governance in Indonesia, aligning with Green AI principles and supporting the democratization of climate intelligence in low-connectivity settings. AbstrakPerubahan iklim menghadirkan tantangan kompleks di Indonesia, khususnya bagi pemerintah daerah dan komunitas lokal yang sering mengalami asimetri informasi dan keterbatasan akses terhadap pengetahuan pakar di wilayah dengan sumber daya dan konektivitas terbatas. Penelitian ini bertujuan mengembangkan dan mengevaluasi Large Language Model (LLM) yang dilokalkan dan diadaptasi ke domain iklim sebagai infrastruktur pengetahuan iklim berbasis offline untuk pendidikan dan tata kelola lokal di Indonesia. Metode penelitian ini menggunakan pendekatan design science research yang meliputi (1) pemilihan Qwen3-4B sebagai base model, (2) kurasi korpus iklim dan transisi energi Indonesia berisi sekitar 12.400 pasangan instruksi-respons (~38 MB) dari dokumen kebijakan iklim nasional, kerangka NDC/RPJMN, panduan energi terbarukan, serta teks ilmiah iklim, (3) parameter-efficient fine-tuning berbasis QLoRA (LoRA rank r=16, alpha=32, learning rate 2e-4, 3 epoch, batch size 2 per perangkat dengan gradient accumulation 4, dan kuantisasi 4-bit NF4), dan (4) deployment offline pada perangkat keras kelas konsumen dengan evaluasi berorientasi tugas terhadap tiga baseline (Qwen3-4B-Thinking, Gemma-3-4B, LLaMa-3.1-8B). Hasil penelitian ini menunjukkan model hasil fine-tuning (Qwen3-4B-REnewbie v1) menghasilkan penurunan perplexity sebesar 15,4% pada data uji domain dan skor kualitatif rata-rata 9,3/10 pada dimensi akurasi faktual, struktur penalaran, dan kepatuhan Bahasa Indonesia, mengungguli seluruh baseline (kisaran 7,0–8,2), serta beroperasi sepenuhnya secara offline pada perangkat konsumen dengan latensi inferensi yang dapat diterima. Kesimpulan yang diperoleh dari penelitian ini adalah LLM yang dilokalkan dan hemat sumber daya dapat berfungsi sebagai infrastruktur pengetahuan iklim yang praktis bagi pendidikan vokasi dan tata kelola lokal di Indonesia, selaras dengan prinsip Green AI dan mendukung demokratisasi kecerdasan iklim di wilayah berkonektivitas terbatas.
Comparison of Coffee Bean Sales Predictions at the Ketiara Coffee Traders Cooperative (KOPEPI) Using Linear Regression and Random Forest Methods Nada Syadzwina; Defry Hamdhana; Ar Razi
Journal of Advanced Computer Knowledge and Algorithms Vol. 3 No. 2 (2026): Journal of Advanced Computer Knowledge and Algorithms - April 2026
Publisher : Department of Informatics, Universitas Malikussaleh

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29103/jacka.v3i2.25974

Abstract

Most of Indonesia's land is used for agriculture and plantations because it is an agrarian country. Harvests or agricultural products can be exported to help the country's economic recovery. Coffee, the most traded tropical crop in the world, is one of the most valuable commodities. Approximately 25 million farming households contribute up to 80% of global coffee production (FAO Organizational 2023). Indonesia's coffee industry continues to experience significant annual growth. To optimize their production and distribution, Indonesian coffee producers must understand coffee bean sales trends. This study compares two methods for predicting coffee bean sales at the KOPEPI Ketiara Aceh Tengah Cooperative using Linear Regression and Random Forest methods. The research methods used in this study are data collection and system design. The results show a comparison of the Linear Regression and Random Forest methods in predicting coffee bean sales. Linear regression provides fairly good accuracy for the price variable with low MAPE values (3.35%–4.55%) and MAE that is still within reasonable limits, but produces large prediction errors for the export variable with high MAPE (67.84%–80.65%) and large MAE (5982–7960). In contrast, Random Forest shows superior performance with very low MAPE (2.69%–3.46%) and smaller MAE (4275–6038) on price variables, as well as more stable and consistent export predictions even though the MAPE values are still quite high (54.25%–84.97%). Overall, Random Forest is a more appropriate model to use because it provides accurate price predictions and more consistent export performance compared to Linear Regression.
A LOW CODE APPROACH TO Q&A ON CARE RECORDS USING FLOWISE AI WITH LLM INTEGRATION AND RAG METHOD Defry Hamdhana
JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika) Vol 9, No 4 (2024)
Publisher : STKIP PGRI Tulungagung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29100/jipi.v9i4.6978

Abstract

Care records are vital for monitoring patient conditions and supporting clinical decision-making, but their diverse formats—such as tables, narrative sentences, checklists, and fill-in-the-blank fields—present challenges for efficient information retrieval. Traditional retrieval methods are often time-consuming and error-prone, while automated systems struggle with contextual accuracy in complex medical language. This study proposes a low-code approach to develop a question-and-answer (QA) system for care records using Flowise AI integrated with Retrieval-Augmented Generation (RAG) methodology. By utilizing LangChain and OpenAI’s language models, Flowise AI provides a framework for constructing a QA system that retrieves information accurately across different documentation formats. The system employs components such as Recursive Character Text Splitter, PDF processing, OpenAI Embeddings, In-Memory Vector Store, and a Conversational Retrieval QA Chain, ensuring efficient retrieval with contextual relevance. Our results demonstrate high accuracy in aligning the QA responses with ground truth data, validating the system's effectiveness in healthcare documentation retrieval. This low-code solution not only enhances accessibility for non-technical users but also empowers healthcare professionals with a scalable tool for quick access to critical patient data. The findings underscore the potential of low-code AI systems like Flowise AI, utilizing RAG, to improve information retrieval in healthcare, supporting more accurate and timely clinical decisions.
Analisis Prediktif Intensi Berwirausaha Mahasiswa Akuntansi Menggunakan Machine Learning Defry Hamdhana; Muhammad Yusra
JURNAL INFORMATIKA DAN KOMPUTER Vol 9, No 2 (2025): Juni 2025
Publisher : Lembaga Penelitian dan Pengabdian Masyarakat - Universitas Teknologi Digital Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26798/jiko.v9i2.1863

Abstract

Penelitian ini bertujuan untuk menganalisis faktor-faktor yang memengaruhi intensi berwirausaha di kalangan mahasiswa akuntansi melalui penerapan model machine learning, khususnya K-Nearest Neighbor (K-NN). Kewirausahaan dianggap memiliki peran penting dalam pertumbuhan ekonomi dan penciptaan lapangan kerja, terutama di negara berkembang seperti Indonesia. Namun, tidak semua mahasiswa menunjukkan minat yang kuat untuk menjadi wirausahawan setelah lulus, termasuk mahasiswa akuntansi yang umumnya memiliki prospek karier di bidang keuangan. Penelitian ini menggunakan Theory of Planned Behavior (TPB) sebagai dasar teoretis untuk memahami faktor sikap, norma subjektif, dan kontrol perilaku dalam memengaruhi intensi berwirausaha mahasiswa. Data dikumpulkan dari 30 mahasiswa akuntansi di Politeknik Negeri Lhokseumawe dan Universitas Islam Kebangsaan Indonesia melalui kuesioner terkait pelatihan kewirausahaan, kemudian dianalisis menggunakan model K-NN. Hasil penelitian menunjukkan bahwa sikap positif dan dukungan sosial memiliki pengaruh signifikan terhadap intensi berwirausaha. Model K-NN dengan parameter K = 3 menunjukkan akurasi sebesar 83%, yang mengindikasikan potensi penerapan machine learning dalam memprediksi intensi berwirausaha. Temuan ini berkontribusi pada literatur kewirausahaan serta memberikan rekomendasi untuk pengembangan program pelatihan kewirausahaan yang lebih komprehensif di lingkungan pendidikan.
Comparative Analysis of XGBoost and CatBoost for Detecting AI-Generated Synthetic Faces Using Texture and Color Features indra firmansyah; Defry Hamdhana; Wahyu Fuadi
Journal of Deep Learning, Computer Vision, and Digital Image Processing Volume 4 Issue 3 September 2026
Publisher : CV. Sakura Digital Nusantara

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.61255/decoding.v4i3.1677

Abstract

Purpose – The rapid advancement of Generative AI produces near-perfect synthetic face images, posing fraud risks. This study compares XGBoost and CatBoost for detecting real versus synthetic faces using texture and color features.Methods – Texture features were extracted using GLCM with four orientations (0°, 45°, 90°, 135°), generating 16 features, while YCbCr mean and standard deviation provided 6 color features (22 total). A balanced 750-image dataset (real and AI from GPT Image 2, Nano Banana, Leonardo AI, Canva, Dreamina AI) was split 80:20 and normalized with StandardScaler.Findings – On an asymmetric, source-shifted test split, where real test images were captured independently while synthetic test images came from the same generators as training, CatBoost outperformed XGBoost, achieving 84.67% accuracy (vs. 80.67%), 83.33% precision (vs. 78.05%), 86.67% recall (vs. 85.33%), and 84.97% F1-score (vs. 81.53%), reducing False Positives by 27.78% and False Negatives by 9.09%. However, repeated 5-fold stratified cross-validation (5×10 repeats) revealed only marginal, non-significant differences (0.05-0.87 percentage points, p > 0.09). Research implications – Although the moderate dataset excludes recent architectures such as StyleGAN, findings affirm the feasibility of efficient, interpretable detection through feature engineering. Since within-domain cross-validation shows no significant algorithmic difference, the observed cross-domain gap should be regarded as preliminary, pending validation through repeated cross-domain evaluations across diverse AI platforms.Originality – The novelty lies in comparing XGBoost and CatBoost on multi-orientation GLCM and YCbCr feature fusion for contemporary AI-generated images, identifying cr_std and contrast_0 as stable, interpretable discriminative markers.
Co-Authors Ar Razi Ar Razi Arliansyah, Arliansyah Baharuddin Ismail Dahlan Abdullah Fadhilah Fadhilah Fadlisyah Fadlisyah Fajriana, Fajriana Fanny Nailufar Fikriah Nst, Nahdah Fitri Maghfirah Hamzani Hamzani indra firmansyah Juanda Pratama Kembaren, Emmia Tambarta Khaidar, Al Lailatul Husna Maizuar Maizuar Mardiaton, Mardiaton Martin Martin Martin Martin Maulidawati, Maulidawati Mauliza Mauliza Mauliza, Mauliza Muhammad Fikry Muhammad Iqbal Muhammad Johan Setiawan Mukti Qamal Mukti Qamal Munirul Ula Nada Syadzwina Nur Afni Yunita Nurdin Nurdin Pangestu, Aridho Putri, Rany Gesta Rio Pratomo Rizal Rizal Rizal Rizal Rizal Rizki Suwanda Safira Adani Sari, Cut Putri Mellita Sayed Fachrurrazi Sri Mulyati Sri Mulyati Suryana, Fitra Taufiq Taufiq Taufiq Taufiq Utomo, Muhammad Fikri Wahyu Fuadi Widyana Verawaty Siregar Yusra, Muhammad Zara Yunizar Zulfikar Aji Kusworo