cover
Contact Name
Muhammad Nur Akbar
Contact Email
muhammad.akbar@uin-alauddin.ac.id
Phone
+6285242044250
Journal Mail Official
instek@uin-alauddin.ac.id
Editorial Address
Jl. H. M. Yasin Limpo No. 36 Samata, Gowa, Sulawesi Selatan
Location
Kab. gowa,
Sulawesi selatan
INDONESIA
JURNAL INSTEK (Informatika Sains dan Teknologi)
ISSN : 25411179     EISSN : 25811711     DOI : https://doi.org/10.24252/instek.v6i2
The Scope topics include, but are not limited to : Agent System and Multi-Agent Systems Analysis & Design of Information System Artificial Intelligence Big Data and Data Mining Cloud & Grid Computing Computer Vision Cryptography Decision Support System DNA Computing E-Government E-Business E-Learning Embedded System Enterprise System Green software development Green computing Green Information Technology Human Computer Interaction Image Processing & Computer Vision Informatics Theory Information System IT for Education IT for Industry IT for Chemical Mechatronics Mobile Computing & Applications Natural Language Processing Network & Data Communications Open Source System Semantic Web Social Networking & Application Soft Computing Software Engineering Software Entrepreneurship Web Engineering Wireless Communication Mikrokontroller elektronic and telecomonunication
Articles 367 Documents
Implementasi Metode Hybrid Retrieval dan Token-Oriented Object Notation pada Chatbot Akademik Berbasis Large Language Model Bayu Winata; Defry Hamdhana; Rizki Suwanda; Fadlisyah; Rini Meiyanti
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.70609

Abstract

Layanan informasi akademik di perguruan tinggi masih sering bersifat konvensional, mengharuskan mahasiswa datang ke kampus atau menghubungi pihak universitas secara manual. Penelitian ini mengembangkan chatbot akademik berbasis Large Language Model (LLM) dengan hybrid retrieval yang menggabungkan data tidak terstruktur dari dokumen akademik dan data terstruktur dari basis data relasional, serta mengimplementasikan Token-Oriented Object Notation (TOON) untuk efisiensi token. Model LLM yang digunakan adalah meta-llama/llama-4-scout-17b-16e-instruct melalui Groq, sedangkan model embedding yang digunakan adalah paraphrase-multilingual-MiniLM-L12-v2. Sistem dibangun menggunakan arsitektur Retrieval Augmented Generation (RAG) dengan ChromaDB dan Text-to-SQL, diintegrasikan melalui Intent Router berbasis prompt-based classification. Pengujian menggunakan metrik ROUGE, Execution Accuracy, dan Token Reduction Rate. Hasil pengujian menunjukkan rata-rata F1-Score sebesar 79,13% pada ROUGE-1 69,01% pada ROUGE-2, dan 78,50% pada ROUGE-L untuk jalur RAG, serta Execution Accuracy 100% pada jalur data terstruktur. TOON berhasil menghemat token rata-rata 24,60% dibandingkan JSON. Chatbot berbasis LLM dengan hybrid retrieval dan TOON mampu memberikan jawaban cepat, tepat, kontekstual, dan efisien dalam penggunaan token.
Position and Language as Diagnostic Variables: A Controlled Invariance Test of Sentiment Classifiers on Indonesian-English Code-Switched Reviews Lie, Kennedy; Santoso, Handri
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.71617

Abstract

Sentiment classifiers are usually judged by accuracy or F1-score, which cannot show whether a model reads the sentiment cue or relies on the surrounding text. Code-switched text, which mixes languages within a sentence, makes this harder to judge. This study proposes a controlled behavioural invariance test in which one sentiment cue is moved across three positions (beginning, middle, end) and written in two languages (Indonesian, English) in 272 reviews, giving 1,632 minimal-pair variants. Logistic Regression, Linear SVC, BERT, mBERT, XLM-RoBERTa and IndoBERT were compared using Prediction Consistency Rate, Prediction Flip Rate and McNemar tests with Holm correction, alongside standard metrics. The two TF-IDF models were largely insensitive to cue position and language but still flipped 5.1% and 1.5% of predictions, and deleting the frequent token “aplikasi” changed their output margins nearly as much as deleting the sentiment cue. BERT and mBERT made more errors on Indonesian than on English cues (adjusted p = 0.014 and 0.031). XLM-RoBERTa was unaffected by the token on the reported split and, over five repeated splits, had the highest mean accuracy and PCR and the lowest PFR. On this dataset, strong standard metrics did not guarantee consistent predictions when the cue was moved or translated.
AI-Based Smart Farming Implementation: A Comparative Analysis of EfficientNet-B0-SVM and ResNet-50-SVM for Cacao Leaf Disease Detection Nurhikma Arifin; Milasari
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.71678

Abstract

Cacao productivity in West Sulawesi remains suboptimal, primarily due to leaf diseases such as Vascular Streak Dieback (VSD) and Leaf Blight, which negatively affect plant growth. Manual disease identification performed by farmers still has several limitations, particularly in terms of subjectivity and accuracy. These limitations become more apparent when distinguishing early-stage disease symptoms that share similar visual characteristics, often leading to misidentification and reduced crop quality. Therefore, this study aimed to develop a more accurate and objective system for cacao leaf disease classification. The dataset consisted of 549 images divided into three classes: Leaf Blight, Healthy, and Vascular Streak Dieback (VSD). The training process was conducted in two stages, namely transfer learning and fine-tuning, while Grid Search was employed to optimize the SVM hyperparameters. The results showed that the EfficientNet-B0-SVM model achieved a validation accuracy of 95.45% and a final test accuracy of 91.82%, whereas the ResNet-50-SVM model achieved a validation accuracy of 93.18% and a final test accuracy of 92.73%. These findings indicate that both models were able to classify cacao leaf diseases with a high level of accuracy. In addition, the application of Principal Component Analysis (PCA) significantly reduced the number of extracted features while preserving more than 95% of the original information. Therefore, the proposed CNN-PCA-SVM approach can be considered an effective method for supporting image-based cacao leaf disease identification.
An Empirical Analysis of Anchor Optimization and RPN Sampling in Faster R-CNN for Pleural Effusion Detection Sakone, Muhammad Jafar; Muhimmah, Izzati
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.72491

Abstract

Pleural effusion detection from chest X-ray images remains challenging due to variations in lesion scale and severe foreground-background imbalance within two-stage object detectors. This study empirically investigates the effects of anchor representation and Region Proposal Network (RPN) sampling on the detection process of Faster R-CNN for pleural effusion detection. A Faster R-CNN with ResNet-50 and Feature Pyramid Network (FPN) was evaluated using representative anchors derived from ground-truth object characteristics and adjusted RPN Intersection over Union (IoU) assignment thresholds. The analysis examined anchor representation, RPN sample distribution, proposal quality, detection confidence, and stage-wise detection outcomes. Representative anchors improved geometric correspondence with the target objects, while RPN threshold adjustment substantially increased positive RPN samples from approximately 4 to 110 per image. Proposal analysis showed high spatial accessibility, with proposal IoU coverage at IoU ≥ 0.50 reaching up to 99.35% across the evaluated configurations. However, stage-wise analysis showed that classification was the dominant source of detection failure, accounting for 15.03–16.99% of the evaluated objects, while proposal and localization failures remained comparatively limited. These findings demonstrate that optimization in anchor representation and RPN sampling can enhance upstream proposal characteristics, but such improvements do not necessarily translate proportionally into final detection performance.
YOLO11m-Based Vehicle Counting for Standardized CO₂ and CO Emission Estimation from Semarang CCTV Setiawan, Moch Adi; Saraswati, Galuh Wilujeng; Lutfina, Erba; Ratmana, Danny Oka; Caturkusuma, Resha Meiranadi
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.73033

Abstract

This study evaluated YOLO11m-based vehicle counting in Pantau Semar CCTV recordings and analyzed how class-specific counting differences affect standardized CO₂ and CO estimates. The dataset comprised 12,000 images: 9,120 for training, 2,280 for validation, and 600 for testing. Model development used main training and corrective fine-tuning with Ultralytics. A Python pipeline combined ByteTrack, class stabilization, and line-crossing counting. Automated counts were compared with manual reference counts across six daytime and nighttime recordings from three locations. The detector achieved precision of 0.904, recall of 0.895, mAP₅₀ of 0.947, and mAP₅₀–₉₅ of 0.833. The system counted 7,444 vehicles compared with 7,667 manually, a difference of −2.91%; daytime and nighttime differences were +0.88% and −6.55%, respectively. Using identical emission factors and a one-kilometer reference distance, system-derived estimates were 1.70–3.88% lower for CO₂ and 2.79–4.94% lower for CO across six scenarios. The study shows how class-specific counting errors and emission factors produce emission-estimate differences that cannot be inferred from total-count differences alone. Findings are limited to the evaluated recordings and assumptions; the results are standardized comparative estimates rather than measurements of actual emissions.
Aplikasi Klasifikasi Jenis Jerawat Menggunakan Algoritma Convolutional Neural Network (CNN) Berbasis Transfer Learning untuk Rekomendasi Penanganan dan Perawatan Kulit Muyassaroh, Siti; Widodo, Tri
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.73149

Abstract

Jerawat (acne vulgaris) merupakan masalah kulit yang umum dan dapat memengaruhi kondisi psikologis penderitanya. Identifikasi jenis jerawat yang masih dilakukan secara visual dan mandiri bersifat subjektif serta berisiko menimbulkan pemilihan perawatan yang kurang tepat. Penelitian ini bertujuan mengembangkan aplikasi web untuk mengklasifikasikan lima jenis jerawat, yaitu blackheads, cyst, papules, pustules, dan whiteheads, menggunakan Convolutional Neural Network (CNN) berbasis transfer learning dengan arsitektur EfficientNetB1, dilengkapi rekomendasi penanganan dan perawatan kulit. Dataset publik dari Kaggle dan Roboflow dibagi menjadi 3.066 data latih, 928 data validasi, dan 927 data uji. Citra diubah ukurannya menjadi 224×224 piksel, kemudian model dilatih menggunakan optimizer Adam dengan variasi jumlah epoch dan learning rate. Model diintegrasikan ke dalam aplikasi web yang menerima citra dari galeri perangkat maupun hasil pengambilan gambar langsung melalui kamera. Pengujian menggunakan tiga seed menghasilkan rata-rata accuracy sebesar 88,35 ± 2,30%, precision makro 89,44 ± 2,42%, recall makro 86,68 ± 3,32%, dan F1-score makro 87,78 ± 3,06%. Untuk keperluan integrasi ke dalam aplikasi web yang membutuhkan satu model definitif, model dari iterasi seed 42 dipilih karena menunjukkan performa evaluasi tertinggi dengan accuracy mencapai 90,83%. Kesalahan klasifikasi paling sering terjadi antara papules dan pustules yang memiliki kemiripan visual. Aplikasi menampilkan jenis jerawat, confidence score, dan rekomendasi perawatan secara otomatis sehingga berpotensi membantu identifikasi awal yang lebih cepat dan objektif.
Ekstraksi Relasi Antar Karakter Pada Teks Satua Bali Menggunakan Support Vector Machine dengan Pendekatan Fitur Leksikon I Kadek Dwi Adnyana; Ngurah Agus Sanjaya ER; I Ketut Gede Suhartana
Jurnal INSTEK (Informatika Sains dan Teknologi) Vol 11 No 2 (2026): OCTOBER
Publisher : Department of Informatics Engineering, Faculty of Science and Technology, Universitas Islam Negeri Alauddin, Makassar, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.24252/instek.v11i2.73157

Abstract

Pemetaan relasi sosial antar tokoh pada Satua Bali penting untuk mendukung preservasi konteks budaya dan pengembangan knowledge graph sastra Nusantara. Penelitian ini mengklasifikasikan relasi Keluarga, Teman, dan Musuh menggunakan Support Vector Machine (SVM) dengan arsitektur Dual-Stream Pipeline yang menggabungkan TF-IDF N-gram pada teks dengan entitas tersamarkan dan fitur leksikon pada teks asli. Teks mentah dikumpulkan dari situs web publik yang memuat cerita Satua Bali, sedangkan kontribusi data penelitian terletak pada penyusunan skema dan anotasi manual tiga kelas relasi. Dari 637 segmen teranotasi, deduplikasi menghasilkan 600 segmen unik yang dievaluasi menggunakan Stratified 5-Fold Cross Validation. Model Dual-Stream mencapai akurasi 75,0% dan Macro F1 75,2%, lebih tinggi dibandingkan baseline TF-IDF pada teks dengan entitas tersamarkan dengan akurasi dan Macro F1 sebesar 71,5%. Sementara itu, TF-IDF pada teks asli mencapai akurasi 83,7%, menunjukkan bahwa nama tokoh dan leksikon asli sangat prediktif, tetapi berpotensi membawa bias identitas. Hasil ini menunjukkan bahwa Dual-Stream mampu memulihkan sebagian informasi relasional akibat masking dan berfungsi sebagai strategi mitigasi terhadap trade-off entity masking.