Jurnal Teknologi Informasi, Komputer, dan Aplikasinya (JTIKA )

EYE DISEASE CLASSIFICATION USING DEEP LEARNING: A COMPARATIVE STUDY OF MOBILENETV2, XCEPTION, AND EFFICIENTNET-B0

Latifa Zahra Agustini, Fitri Bimantoro, Ramaditia Dwiyansaputra — Mon, 30 Mar 2026 00:00:00 +0800

This study presents a comparative analysis of three convolutional neural network (CNN) architectures—MobileNetV2, Xception, and EfficientNet-B0—for classifying retinal fundus images into four categories: Cataract, Diabetic Retinopathy, Glaucoma, and Normal. Using a dataset of 4,217 images, the models were trained with transfer learning, image augmentation, and regularization techniques, and evaluated through 5-fold cross-validation. EfficientNet-B0 achieved the highest mean accuracy (0.85) and demonstrated stable performance across all metrics, while MobileNetV2 provided competitive accuracy with lower computational requirements, making it suitable for resource-limited environments. Xception showed the lowest and least stable performance, indicating a higher tendency to overfit. External validation with clinical images revealed a significant drop in accuracy for all models, highlighting challenges related to domain shift and limited generalization. Grad-CAM analysis also showed difficulties in detecting subtle pathological features in Diabetic Retinopathy and Glaucoma. The study is limited by the small dataset size, reliance on a single data source, and the absence of additional clinical information. Future work should incorporate larger and more diverse datasets, apply domain adaptation strategies, and integrate multimodal clinical data to enhance robustness and clinical applicability.

PENGEMBANGAN SISTEM INFORMASI MONITORING PRAKTIK KERJA LAPANGAN BBPOM MATARAM DENGAN METODE RAPID APPLICATION DEVELOPMENT

Bagas Adinata, Nadiyasari Agitha, Ida Bagus Ketut Widiartha — Tue, 31 Mar 2026 00:00:00 +0800

The Mataram National Agency of Drug and Food Control (BBPOM Mataram) requires an information system to manage internship participants’ data more effectively. The administrative process has been carried out manually, which often leads to inefficiency and potential errors in data storage. This study aims to develop a web-based information system that integrates the entire internship process, including application submission, entrance selection through pre-test, attendance recording, daily activity documentation, and final score calculation based on attendance and activities. The system was developed using the Rapid Application Development (RAD) method, which emphasizes iterative prototyping and user feedback to accelerate the development process. System testing was conducted using the black-box testing method to evaluate the functionality of each feature according to the requirements. In addition, usability testing was performed using the System Usability Scale (SUS), resulting in a score of 79,16, which indicates that the system is considered Good in terms of usability. The test results show that all system features function properly and meet the specified requirements. The implementation of this system supports BBPOM Mataram in improving efficiency, accuracy, and integration in managing internship participant data.

PENGEMBANGAN SISTEM INFORMASI PORTOFOLIO PEMBELAJARAN UNTUK MENDUKUNG MANAJEMAN CAPAIAN PEMBELAJARAN MATAKULIAH BERBASIS OBE (STUDI KASUS DI PROGRAM STUDI TEKNIK INFORMATIKA)

Tue, 31 Mar 2026 00:00:00 +0800

The Information Technology Study Program (PSTI) implements an OBE-based curriculum to measure graduate achievement (CPL) and course learning achievement (CPMK), requiring lecturers to compile learning achievement reports in the form of portfolios at the end of each semester. The compilation of portfolios is supplemented with CPL and CPMK results calculated using Microsoft Excel. During the preparation of the report, difficulties arise when changing data that affects the formula. Not all lecturers understand the available formulas and equations. The purpose of this research is to develop an information system to support the preparation of lecturers' course learning portfolio reports. The method used is Extreme Programming, which is an Agile software development approach with the stages of Planning, Design, Coding, Testing, and Release Phase (Deploy). This research resulted in a Learning Portfolio Information System to Support OBE-Based Course Learning Achievement Management (Case Study in the Informatics Engineering Study Program). In this information system, lecturers can manage CPL, CPMK, Sub_CPMK, Assessment, Evaluation, Results, and Portfolio data. From testing using User Acceptance Testing, the results obtained were an average of 37% strongly agree, 43% agree, and 19% somewhat agree. Therefore, it can be said that the information system created can be used properly.

KLASIFIKASI PENYAKIT PNEUMONIA PADA X-RAY PARU-PARU MENGGUNAKAN MODEL HYBRID GRAY LEVEL CO-OCCURRENCE MATRIX DAN ARTIFICIAL NEURAL NETWORK

Amdila Rahmadi, I Gede Pasek Suta Wijaya, Pahrul Irfan — Tue, 31 Mar 2026 00:00:00 +0800

Pneumonia is a leading cause of morbidity and mortality, particularly in children, requiring rapid and accurate diagnosis. This study proposes a hybrid classification model that combines Gray Level Co-occurrence Matrix (GLCM) texture feature extraction with an Artificial Neural Network (ANN) to analyze chest X-ray images. The dataset consisted of 3,150 images, balanced using random undersampling. GLCM features were extracted across multiple distances and four orientations, generating 19 texture features per image. Seven experimental scenarios were conducted to evaluate ANN architectures with 2, 3, and 4 fully connected layers to identify the most effective configuration. The best-performing model achieved an accuracy of 91.50%, with precision, recall, and F1-score of 0.91, demonstrating consistent performance in distinguishing normal and pneumonia cases. Due to its relatively low computational complexity, this approach is suitable for low-resource healthcare settings. Future work will focus on expanding the dataset and validating the model with clinical data to enhance real-world applicability.

PEMBELAJARAN INTERAKTIF SISTEM TATA SURYA MELALUI AUGMENTED REALITY DALAM PENDIDIKAN DASAR

Awang Pradana, Radiansyah Radiansyah — Tue, 31 Mar 2026 00:00:00 +0800

This study aims to design and develop interactive learning media based on Augmented Reality (AR) for solar system material, targeting sixth-grade students at SDN 012 Tarakan. The media development was carried out using the ADDIE model, which includes the stages of analysis, design, development, implementation, and evaluation. This media was created using the Unity and Vuforia SDK platforms, equipped with 3D objects and audio explanations to aid student understanding. The study involved 26 sixth-grade students through purposive sampling. The material presented included an introduction to the eight planets, the Sun, and the Moon. whereas the functionality test (Blackbox Testing) revealed that 16 application features functioned well. The media expert validation results obtained an average score of 94%, and the material expert validation reached 81.66%, both of which were in the very feasible category. Student responses were also very positive, with an average of 96.09%. Thus, this AR-based learning media is deemed effective, engaging, and suitable for use in the learning process at SDN 012 Tarakan.

PERBANDINGAN SUMBER DAYA HONEYPOT BERBASIS COWRIE DAN OPENCANARY TERHADAP SERANGAN SSH BRUTE-FORCE DAN PORT-SCANNING

Yoga Pramana, Raphael Bianco Huwae, Andy Hidayat Jatmika — Tue, 31 Mar 2026 00:00:00 +0800

The rapid advancement of information technology demands good security. One approach that can be taken and is considered effective for identifying and analyzing cyberattacks is to implement a honeypot security layer. In a comparative study of two types of honeypots, namely Opencanary and Cowrie, which examined the form of honeypot response and the main focus of resource requirements to attacks along with the use of two methods of brute-force and port-scanning attacks. The attack was carried out virtually with the self-attack method, with the aim of comparing each honeypot in terms of resource requirements. The results show that in brute-force attacks, Opencanary has lower resource requirements with the highest CPU/RAM requirements of only 14% / 1.3%, while Cowrie requires more resources with the highest CPU/RAM requirements of 17% / 1.3%. While port-scanning attacks have lower resource requirements with the highest CPU requirements in Opencanary at 3% and Cowrie at 2% and similar RAM requirements at 1.28%. This study is expected to be a benchmark in selecting a honeypot that is tailored to the existing resource requirements.

SISTEM DETEKSI BERITA PALSU DUA BAHASA MENGGUNAKAN TF-IDF DAN MULTINOMIAL NAIVE BAYES

Rheno Septianto, Yan Rianto — Tue, 31 Mar 2026 00:00:00 +0800

The rapid spread of misinformation poses a major threat to public trust and digital literacy. This study develops a bilingual fake news detection system capable of analyzing news content in English and Indonesian. The system uses two separate monolingual models trained independently on the WELFake dataset (English) and the Berita Hoax 2023 dataset (Indonesian). Each model applies text preprocessing techniques such as tokenization, stopword removal, and normalization before transforming the text using TF-IDF. The classification process utilizes the Multinomial Naïve Bayes algorithm, chosen for its efficiency in handling high-dimensional text data. The bilingual system integrates an automatic language detection module that selects the appropriate model based on the detected language. Evaluation results show that the English model achieves an accuracy of 86%, while the Indonesian model achieves an accuracy of 93%. These results indicate that the two-model bilingual approach provides reliable performance for multilingual fake news detection. This study contributes to practical solutions for misinformation mitigation, especially in multilingual environments like Indonesia.

PENERAPAN DATA MINING MENGGUNAKAN ALGORITMA K-MEANS UNTUK ANALISIS DATA BELANJA ONLINE MAHASISWA

Deiva Verlyn Marjuki, Mutyara Safitri, Harun Al Rosyid — Tue, 31 Mar 2026 00:00:00 +0800

This study was conducted to analyze the online shopping behavior of college students using the K-Means algorithm as a clustering technique in data mining. This study was motivated by the lack of systematic segmentation of student shopping behavior, which limits the understanding of purchasing characteristics within this consumer group. Unlike previous studies that mostly examine general retail customers or broad e-commerce users, this study specifically focuses on university students by integrating demographic and behavioral attributes. The originality of this study is reflected in the simultaneous use of six variables, namely gender, shopping time, product type, expenditure level, payment method, and purchase decision factors. Data were collected through an online survey involving 200 active college students. The research stages consisted of data cleaning, data category transformation using One-Hot Encoding, clustering model construction using the K-Means algorithm, and cluster evaluation using the Silhouette method. The evaluation results showed that the optimal number of clusters was k = 3, achieving the of 0.0913. Three distinct segments of college students' online shopping behavior were identified, providing insights that can support more targeted marketing strategies and student-oriented e-commerce services.

ANALISIS KESESUAIAN PENGUKURAN KALORI SMARTWATCH DENGAN PERHITUNGAN MET PADA AKTIVITAS OLAHRAGA

Rafli Assiddiqie Raihan, Irving Vitra Paputungan, Mukhammad Andri Setiawan — Tue, 31 Mar 2026 00:00:00 +0800

This study aims to evaluate the agreement between calorie estimations generated by smartwatches and manual calculations based on the Metabolic Equivalent of Task (MET) during physical activity. Three participants with different physiological characteristics and activity intensities completed 15 training sessions using the Xiaomi Smart Band 8 and 10. Calorie estimates from the devices were compared with MET-based calculations using the paired sample t-test. The results indicate that, for moderate to high intensity activities such as jogging and running, no significant differences (p > 0.05) were observed between the two estimation methods, suggesting a good level of agreement. Conversely, low-intensity walking showed significant differences (p < 0.05), reflecting a tendency for overestimation by the smartwatch. Overall, the agreement improved when heart rate rhythm and movement patterns were more stable, consistent with physiological principles relating oxygen consumption and MET values. As a preliminary case-series, this study highlights the importance of activity intensity when interpreting smartwatch-based energy estimates and provides insight into the practical use of wearable devices for daily exercise monitoring.

PYTHON WEB SYSTEM TO RESTORE SQL SERVER DATABASE TO DRC WITH ADVANCED INFORMATION RETRIEVAL

Devis Rabertra, Irwansyah Saputra — Tue, 31 Mar 2026 00:00:00 +0800

Disaster Recovery Centers (DRC) play a crucial role in ensuring the availability and continuity of database operations in enterprise environments. The process of restoring databases from production servers to DRCs is often performed manually, which can lead to errors such as selecting incorrect backups, corrupted files, and lengthy search times. The complexity increases with the growing number of databases and the variety of daily backup types.This study develops an automated system based on a Python Web Interface integrated with Advanced Information Retrieval (IR) to improve the accuracy and speed of finding relevant backups before restoration. The system employs Natural Language Processing (NLP) and multi-criteria relevance scoring, evaluating backup suitability based on fuzzy matching of database names, recency, semantic similarity, backup type, and file size.Testing was conducted using 28 backup records from 5 different databases. Results show that Advanced IR can accelerate backup searches in under 2 seconds, with relevance ranking ranging from 38% to 67%. Additionally, the automated restore process via Python achieved an average execution time of 7.49 seconds with a 100% success rate.

VISUALISASI KEBAYA BALI BERBASIS AUGMENTED REALITY UNTUK PENINGKATAN PROMOSI UMKM

I Gusti Ayu Agung Mas Aristamy, Putu Risanti Iswardani, Ni Putu Suci Meinarni — Tue, 31 Mar 2026 00:00:00 +0800

Penjualan produk kebaya Bali pada UMKM HitaFelicite masih menggunakan media promosi konvensional berupa foto katalog dan media sosial dua dimensi sehingga konsumen belum dapat melihat bentuk produk secara detail dan interaktif. Keterbatasan tersebut menyebabkan pengalaman pengguna dalam mengeksplorasi produk menjadi kurang optimal. Oleh karena itu, penelitian ini bertujuan untuk mengembangkan aplikasi katalog produk kebaya Bali berbasis Augmented Reality (AR) sebagai media visualisasi produk yang lebih interaktif. Metode pengembangan aplikasi menggunakan tahapan pemodelan sistem, pembuatan aset 3D, serta implementasi teknologi AR berbasis perangkat mobile. Pengujian sistem dilakukan menggunakan Black Box Testing, pengujian spesifikasi perangkat (device compatibility), dan User Experience Questionnaire (UEQ) untuk mengetahui tingkat keberhasilan fungsi aplikasi dan pengalaman pengguna. Evaluasi pengalaman pengguna menggunakan User Experience Questionnaire (UEQ) terhadap 20 responden menghasilkan penilaian positif, dengan aspek Efisiensi berada pada kategori Excellent, sementara Daya Tarik, Kejelasan, dan Ketepatan pada kategori Good. Dengan demikian, aplikasi AR yang dikembangkan dapat menjadi media promosi digital yang lebih interaktif bagi UMKM HitaFelicite Kebaya.

PERBANDINGAN MODEL DEEP LEARNING UNTUK PENERJEMAHAN BAHASA ISYARAT SIBI BERBASIS MOBILE

I Putu Yoga Indrawan, Christina Purnama Yati, Ni Luh Wiwik Sri Rahayu — Tue, 31 Mar 2026 00:00:00 +0800

Komunikasi antara penyandang tunarungu dan masyarakat umum di Indonesia masih terbatas akibat rendahnya pemahaman terhadap Sistem Bahasa Isyarat Indonesia (SIBI) yang secara resmi diakui oleh pemerintah. Penelitian ini bertujuan untuk mengevaluasi performa tiga model deteksi objek berbasis deep learning MobileNetV2-SSD, MobileNetV2-RetinaNet, dan YOLOv11 dalam mendeteksi bahasa isyarat SIBI, serta merekomendasikan model terbaik untuk implementasi di perangkat Android. Sistem dirancang dengan fokus pada efisiensi agar dapat digunakan secara optimal di perangkat mobile. Hasil evaluasi menunjukkan bahwa MobileNetV2-SSD memberikan performa terbaik dengan mean Average Precision (mAP) sebesar 99,7% dan kecepatan 9 frame per second (FPS). YOLOv11 memperoleh mAP sebesar 89,8% dan 5 FPS, meskipun mengalami fluktuasi pada validation loss. Sementara itu, MobileNetV2-RetinaNet awalnya mencatat mAP sebesar 38,8%, namun meningkat hingga 87,69% pada rasio dataset 70:15:15. Meskipun akurasinya membaik, model ini tetap kurang ideal karena kecepatan inferensi hanya mencapai 2 FPS.Penelitian ini diharapkan dapat menjadi kontribusi awal dalam pengembangan teknologi penerjemah bahasa isyarat yang inklusif dan efisien, guna meningkatkan aksesibilitas komunikasi bagi penyandang tunarungu di Indonesia.

PENGGUNAAN ALGORITMA TEXTRANK UNTUK PERINGKASAN OTOMATIS BERITA BAHASA INDONESIA

I Putu Mahesa Kama Artha, Ni Wayan Jeri Kusuma Dewi, Ni Luh Wiwik Sri Rahayu — Tue, 31 Mar 2026 00:00:00 +0800

Situs berita sering menyajikan paragraf atau kalimat yang panjang, disertai dengan berbagai detail tambahan yang tidak selalu relevan dengan kebutuhan informasi pembaca. Oleh karena itu, diperlukan metode seperti Textrank untuk melakukan peringkasan otomatis. Penelitian ini membahas penerapan algoritma Textrank dalam peringkasan otomatis berita berbahasa Indonesia. Hasil evaluasi menunjukkan nilai ROUGE-1 dengan Precision 0.7708, Recall 0.6930, dan F1-Score 0.7216. ROUGE-2 dengan Precision 0.6936, Recall 0.6212, dan F1-Score 0.6480, serta ROUGE-L dengan Precision 0.7210, Recall 0.6465, dan F1-Score 0.6741. Hasil rata-rata matriks evaluasi ROUGE akhir sebesar 0.6812 mengindikasikan bahwa sistem memiliki performa yang baik dalam menghasilkan ringkasan.

DESIGN AND DEVELOPMENT OF A POSTCARD-INTEGRATED AUGMENTED REALITY APPLICATION FOR EDUCATIONAL EXPLORATION OF LOMBOK ISLAND TOURIST ATTRACTIONS

Tue, 31 Mar 2026 00:00:00 +0800

Indonesia possesses diverse tourism destinations, including Lombok Island, known for its natural landscapes and cultural heritage. However, the introduction of local tourism knowledge to elementary school students remains limited, as textbooks and two-dimensional images predominate, which are less engaging for digital-age learners. This study aims to design and develop an augmented reality-based mobile application to support the educational exploration of tourist attractions on Lombok Island. The development process adopted the Multimedia Development Life Cycle model, which comprises the concept, design, material collection, assembly, testing, and distribution stages. The application visualizes prominent destinations such as Mount Rinjani, Gili Trawangan, and Sade Village through interactive three-dimensional objects accompanied by concise educational information. Functional testing results indicate that the application operates effectively on mobile devices and provides an interactive learning experience. The findings suggest that integrating Augmented Reality technology into local geography and cultural education enhances student engagement and understanding. The application can serve as an innovative digital learning medium to promote awareness and appreciation of regional heritage among elementary school students.