Claim Missing Document
Check
Articles

Found 6 Documents
Search

Ketahanan Pembelajaran Mesin terhadap Adversarial examples: Metodologi dan Pertahanan Ade Kurniawan; Ely Aprilia; Achmad Indra Aulia; Amril Mutoi Siregar; Leonard Goeirmanto
Faktor Exacta Vol 18, No 2 (2025)
Publisher : LPPM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30998/faktorexacta.v18i2.26078

Abstract

This paper examines the vulnerability of machine learning models to adversarial examples: inputs that are subtly manipulated to deceive a model into making incorrect predictions. Although deep learning has demonstrated remarkable performance across various tasks, the security of these models remains a significant challenge. This study provides a comprehensive review of various methods for generating adversarial examples, a classification of attack techniques, and corresponding defense strategies, including both active and passive approaches. The findings indicate that a combination of several defense techniques is significantly more effective in enhancing model robustness compared to any single approach. This research is expected to provide a foundation for the development of more secure and reliable machine learning models for critical applications.
AI-AUGMENTED HEALTH INFORMATION SYSTEM DESIGN FOR SCALABLE MOBILE MENTAL WELLNESS SUPPORT IN INDONESIA Tiawan; Eliza Ariesta; Nilam Atsirina Krisnaputri; Amril Mutoi Siregar; Surjandy; Merios Gusan Putra; Dani Lukman Hakim; Samsul Arifin; Nur Dava Kurniawan
Jurnal Informatika Teknologi dan Sains (Jinteks) Vol 8 No 1 (2026): EDISI 27
Publisher : Program Studi Informatika Universitas Teknologi Sumbawa

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.51401/jinteks.v8i1.6974

Abstract

Indonesia faces a youth mental health crisis, with limited access to professional care. WHO recommends one mental health worker per 30.000 people, but Indonesia only has 1:200.000 for general mental health personnel and 1:223.587 for psychiatrists. This paper addresses the implementation gap through the design and integration of an AI augmented health information system within Teman Curhat, a mobile mental wellness platform. We applied thematic analysis to expert interviews and used K-Means clustering on 16.689+ user reported cases to classify concerns into three severity levels. The system’s novelty lies in embedding on device Gemini AI for offline triage, ethical chatbot support, and adaptive user routing. Key outputs include a validated clustering model (Silhouette Score: 0.61), AI assisted classification logic, integration within Android Studio, and a scalable conversational interface. This innovation supports SDG 3 (Good Health and Well being) and SDG 10 (Reduced Inequalities) by enabling inclusive, responsive, and privacy focused digital mental health services for underserved communities.
MODEL CHATBOT AI AGENT BERBASIS CONVERSATIONAL AGENT UNTUK BISNIS DIGITAL LAPTOPCARE Tiawan; Eliza Ariesta; Nilam Atsirina Krisnaputri; Amril Mutoi Siregar; Surjandy; Merios Gusan Putra; Dani Lukman Hakim; Samsul Arifin; Muhammad Fathir Fahlevi; Favian Jarsi Taufiqqurakhman; Hexsel Aldoegasha; Alya Nabilah; Taufiqqurahman Hutri; Nur Davi Kurniawan
Jurnal Informatika Teknologi dan Sains (Jinteks) Vol 8 No 1 (2026): EDISI 27
Publisher : Program Studi Informatika Universitas Teknologi Sumbawa

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.51401/jinteks.v8i1.6975

Abstract

Perkembangan teknologi Conversational Agent (CA) memungkinkan pengembangan chatbot berbasis kecerdasan buatan untuk mendukung layanan bisnis digital. Penelitian ini bertujuan untuk mengembangkan dan mengimplementasikan chatbot AI Agent sebagai layanan dukungan pelanggan pada model bisnis perawatan laptop yang dikembangkan oleh mahasiswa. Metode yang digunakan adalah pendekatan Conversational Agent dengan integrasi n8n sebagai platform otomasi alur kerja, Telegram sebagai saluran komunikasi, Google Gemini Chat Model sebagai pemroses bahasa, serta Google Spreadsheet sebagai sumber data terstruktur untuk informasi produk dan layanan. Sistem chatbot dirancang sebagai agen berbasis tugas dengan aturan bisnis yang mengatur jenis pertanyaan dan layanan yang dapat diberikan. Hasil penelitian menunjukkan bahwa chatbot mampu memberikan respons yang akurat, kontekstual, dan konsisten terkait informasi layanan, stok barang, estimasi harga, serta kebutuhan pelanggan secara real-time. Implementasi ini membuktikan bahwa integrasi AI Agent dengan platform otomasi dan data terstruktur dapat meningkatkan efisiensi layanan pelanggan serta mendukung pengembangan model bisnis digital di bidang perawatan laptop.
PERANCANGAN FITUR DARURAT DAN PENGINGAT PADA APLIKASI JAMAAH UMROH XYZ MENGGUNAKAN SCRUMBAN Tiawan; Nusaibah; Rifky Kurniawan; Eliza Ariesta; Nilam Atsirina Krisnaputri; Amril Mutoi Siregar; Surjandy; Merios Gusan Putra; Dani Lukman Hakim; Samsul Arifin; Citra Nur Napiah
Jurnal Informatika Teknologi dan Sains (Jinteks) Vol 8 No 1 (2026): EDISI 27
Publisher : Program Studi Informatika Universitas Teknologi Sumbawa

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.51401/jinteks.v8i1.7049

Abstract

The problem of Umrah pilgrims getting lost and losing personal belongings is still common, especially in high-density areas such as the Grand Mosque and the Prophet's Mosque, with more than 50 cases of missing pilgrims per day during the peak Umrah season. A survey of 50 pilgrims showed that 86% of respondents experienced location disorientation and 74% had lost or nearly lost personal belongings. This study aims to design emergency and reminder features in the XYZ Umrah Pilgrim Application to improve security and management of pilgrim activities. The development method used is Scrumban, a combination of the iterative and structured Scrum method with the Kanban method that emphasizes workflow visualization and task management execution. The emergency feature is designed using the Fused Location Provider to send pilgrim location coordinates in real-time with a detection time of less than 30 seconds, while the reminder feature helps manage luggage and prayer schedules through a notification system. Functional and performance testing results show a feature success rate above 90%. The conclusion of this study shows that the integration of emergency and reminder features in one Umrah application with the Scrumban approach is able to improve the security and comfort aspects of the congregation.
Enhancing Cow Pose Recognition Through Morphological Feature Extraction and AI Models Amril Mutoi Siregar; Ahmad Indra Aulia; Gunawan Witjaksono; Sony Hartono Wijaya; Ahmad Turmudi Zy
Journal of Embedded Systems, Security and Intelligent Systems Vol 7 No 3 (2026): September 2026
Publisher : Program Studi Teknik Komputer

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59562/jessi.v7i3.12426

Abstract

Purpose – This study investigates how different image segmentation techniques influence machine-learning performance in cow pose classification and identifies an effective combination of preprocessing, segmentation, and classification methods for image-based livestock monitoring. Design/methods/approach – A publicly available dataset comprising 271 two-dimensional RGB cow images was classified into four pose categories: front, left, right, and others. Images were processed using brightness adjustment and Gaussian blur, followed by Canny edge detection, Laplacian filtering, or K-Means clustering. Support Vector Machine (SVM) and Random Forest (RF) classifiers were then evaluated using 10-fold cross-validation. Model performance was assessed using accuracy, precision, recall, F1-score, and receiver operating characteristic–area under the curve (ROC-AUC). Findings – The K-Means-based segmentation configurations consistently outperformed Canny- and Laplacian-based approaches. Random Forest combined with K-Means achieved the best overall performance, with 99.60% accuracy, 98.99% precision, 99.59% recall, and a 98.99% F1-score. The corresponding K-Means-based SVM model achieved 98.16% accuracy. In contrast, Laplacian-based configurations produced the lowest classification accuracy, ranging from approximately 64% to 65%. These findings indicate that segmentation quality substantially affects the discriminative representation available to conventional machine-learning classifiers. Research implications/limitations – The findings demonstrate the potential of combining relatively simple segmentation and machine-learning methods for cow pose recognition when dataset size and computational resources are limited. However, the study is restricted to a relatively small publicly available image dataset and four pose categories; therefore, validation using larger, more heterogeneous, farm-based datasets is required before practical deployment. Originality/value – This study provides a comparative evaluation of multiple classical segmentation techniques under a consistent machine-learning framework and highlights the effectiveness of K-Means segmentation combined with Random Forest for cow pose classification.
Perbandingan Algoritma Random Forest, Logistic Regression, dan Decision Tree Untuk Prediksi  High-Score Anime pada Dataset Myanimelist Muhammad Arfy Rafa Fakhrezie; Amril Mutoi Siregar
Jurnal Riset dan Aplikasi Mahasiswa Informatika (JRAMI) Vol. 7 No. 04 (2026): Jurnal Riset dan Aplikasi Mahasiswa Informatika (JRAMI)
Publisher : Program Studi Teknik Informatika, FTIK, Universitas Indraprasta PGRI

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30998/jrami.v7i04.1269

Abstract

The MyAnimeList platform provides rich structured data encompassing content metadata, user interactions, and community scores. This study aims to compare the performance of three classification algorithms—Random Forest, Logistic Regression, and Decision Tree in predicting whether an anime qualifies as high-score (score ≥ 7.0) based on attributes such as member count, episodes, favorites, popularity rank, type, status, source, and genre. The dataset used is the MyAnimeList Dataset published by azathoth42 on Kaggle, downloaded via KaggleHub. Research stages include exploratory data analysis, feature engineering (label encoding for categorical variables, one-hot encoding for genres, and derived features fav_ratio and is_long_series), numerical feature normalization for Logistic Regression, and stratified 80:20 data splitting. Evaluation uses Accuracy, Precision, Recall, F1-Score, and ROC-AUC metrics, supplemented by 5-fold stratified cross-validation. Results show that Random Forest achieves the best performance, followed by Logistic Regression and Decision Tree. Feature importance analysis reveals that members, rank, and favorites are the most dominant predictors. This study demonstrates that Random Forest ensemble approaches consistently outperform single models for anime content quality classification tasks.