cover
Article Per Year (5 Year)
All
Home Page
OAI Link
Editorial Team
Contact
Reviewer
Google Scholar
Contact Name
Huzain
Contact Email
huzain.azis@umi.ac.id
Phone
+628114484875
Journal Mail Official
ijodas.journal@gmail.com
Editorial Address
Jln. Paccerakkang, Kel. Berua, Kec.Biringkanaya, Kota Makassar, Propinsi Sulawesi Selatan, 90241
Location
Unknown,
Unknown
INDONESIA
Indonesian Journal of Data and Science
Published by yocto brain
ISSN : -     EISSN : 27159930     DOI : -
Core Subject : Science, Education,
Computer Science & IT Decision Sciences, Operations Research & Management Mathematics
IJODAS provides online media to publish scientific articles from research in the field of Data Science, Data Mining, Data Communication, Data Security and Data Representation
Arjuna Subject : Ilmu Komputer (semua kategori) - Ilmu Komputer (Lain-Lain)
Articles 191 Documents
 16   17   18   19   20 
The Effect of Clinical Rule-Based Domain Filtering on the Performance of FP-Growth-Based Drug Recommendation Systems Muhammad Zaqly Luluang; Irawati; Herdianti Darwis
Indonesian Journal of Data and Science Vol. 7 No. 1 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i1.398

Abstract

This study analyzes the effect of domain filtering on drug recommendation systems based on association rule mining using the FP-Growth algorithm with Neural Collaborative Filtering (NCF) as a comparison. The dataset used was derived from patient medical records containing attributes such as complaints, diagnoses, and drug therapies, with a total of 1,000 patient transactions. To avoid data leakage, the dataset was randomly divided into 70% training data and 30% test data before the modeling process was carried out. Domain filtering was applied by limiting the rule structure so that complaints and diagnoses acted as antecedents and drugs as consequents. The performance of the recommendation system was evaluated using the Precision@5, Recall@5, and Normalized Discounted Cumulative Gain (NDCG@5) metrics. The results of the experiment show that the FP-Growth approach with domain filtering produces higher Precision@5 and NDCG@5 values than the non-filtering approach. The Wilcoxon Signed-Rank test shows that the difference is statistically significant, while effect size analysis using Cliff's Delta shows a practically meaningful impact. Furthermore, a comparison with Neural Collaborative Filtering shows that the collaborative filtering-based approach is less effective on transactional clinical prescription data with limited historical interactions. These findings indicate that integrating medical domain knowledge into FP-Growth can improve the clinical relevance and quality of drug recommendation rankings
Website-Based Boycott Product Detection System using Convolutional Neural Network: A comparison of YOLOv8 and VGG16 Mariani Ani; Dolly Indra Indra; Sitti Rahma Jabir Rahma
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.317

Abstract

Introduction: Product boycotts have become a common form of social response in the digital era, yet manual identification of boycotted products can be slow and inaccurate. This study develops a website-based system and compares YOLOv8 and VGG16 for automated identification of boycott and non-boycott products from images. Method: A dataset of 4,250 food and beverage product images, comprising 3,638 boycott and 612 non-boycott products, was collected from internet sources and divided into 70% training, 20% validation, and 10% testing sets. YOLOv8 was trained using 640×640-pixel inputs as a one-stage object detector, while VGG16 used 224×224-pixel inputs with transfer learning as an image classifier. Both models were integrated into a website-based detection system and evaluated using accuracy, precision, recall, and F1-score. Results and Discussion: On 425 test images, YOLOv8 achieved 91.7% accuracy, 99.7% precision, 90.3% recall, and a 94.9% F1-score, substantially outperforming VGG16, which achieved 51.3% accuracy, 81.6% precision, 55.7% recall, and a 66.2% F1-score. YOLOv8 demonstrated greater robustness to variations in background, lighting, and product appearance because of its object-localization capability. Conclusion: YOLOv8 is more effective than VGG16 for website-based boycott product detection and provides a stronger foundation for practical real-time identification systems.
Comparison of ResNet50 and ResNet101 Feature Extraction for Tea Leaf Disease Classification Using Support Vector Machine Wistiani Astuti; Erick Irawadi Alwi; Farniwati Fattah; Tasrif Hasanuddin; Julisa; Ulfa Sari; Ainur Rahma Almagfirah
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.338

Abstract

Introduction: Tea leaf diseases can substantially reduce crop quality and productivity, making early and accurate diagnosis important for effective disease management. This study compares ResNet50 and ResNet101 as pretrained deep feature extractors combined with Support Vector Machine (SVM) to determine whether a deeper residual architecture can improve discrimination among visually similar tea leaf disease classes. Method: Images were obtained from the Kaggle “Identifying Disease in Tea Leaves” dataset comprising eight classes. Data augmentation increased each class to 800 images, yielding 6,400 images that were divided into training and testing sets using an 80:20 ratio. ResNet50 and ResNet101 pretrained on ImageNet were used as fixed feature extractors, and the resulting feature vectors were standardized and classified using an RBF-kernel SVM. Results and Discussion: ResNet101–SVM achieved the best performance with 97.97% accuracy and precision, recall, and F1-score of 98%, substantially outperforming ResNet50–SVM, which achieved 89.15% accuracy, 90% precision, 89% recall, and 89% F1-score. The deeper ResNet101 architecture provided more discriminative representations for visually similar disease patterns, although a small number of misclassifications remained. Conclusion: ResNet101 combined with SVM provides a more accurate and reliable framework than ResNet50–SVM for multi-class tea leaf disease classification and offers a promising foundation for automated disease diagnosis systems.
Applicant Data Segmentation and Pattern Analytics for University Admissions Strategy Using Hybrid SOM and K-Means Diyah Ruswanti; Dahlan Susilo
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.388

Abstract

Introduction: Effective university admissions strategies require a clear understanding of applicant demographics, academic characteristics, geographic origins, and information channels. This study applies a hybrid clustering approach to identify meaningful applicant segments that can support targeted recruitment and resource allocation. Method: Historical applicant records from Universitas Sahid Surakarta covering 2021–2025 were preprocessed through data cleaning, one-hot encoding of categorical variables, and Min-Max normalization. A hybrid Self-Organizing Map (SOM) and K-Means framework was employed, where SOM projected high-dimensional applicant characteristics into a lower-dimensional topological representation and K-Means partitioned the resulting prototypes into distinct clusters. Clustering quality was evaluated using the Davies–Bouldin Index (DBI) to determine the optimal number of segments. Results and Discussion: The lowest DBI was obtained for three clusters, indicating the most appropriate segmentation structure. The resulting groups were characterized as proximity-driven local applicants, regional career-oriented applicants dominated by vocational-school backgrounds, and high-achieving out-of-region applicants with stronger academic performance and greater reliance on institutional websites and search channels. These patterns provide actionable insight for differentiated recruitment strategies. Conclusion: The hybrid SOM–K-Means approach effectively identifies interpretable applicant segments and provides descriptive intelligence that can support more targeted marketing, channel selection, scholarship strategies, and admissions resource allocation in higher education
Hybrid Deep Learning Models For Gold Price Prediction: Enhancing Forecast In Volatile Financial Markets Ni Luh Wiwik Sri Rahayu Ginantra; Ni Wayan Yeni Pratiwi; Christina Purnama Yanti; Wayan Gede Suka Parwita
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.394

Abstract

Introduction: Gold is widely regarded as a long-term store of value and a hedge against inflation, yet its short-term price volatility creates significant challenges for investment decision-making and requires accurate forecasting methods. This study evaluates a hybrid deep learning approach combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to capture both short-term fluctuations and long-term temporal dependencies in gold price movements. Method: Historical daily gold closing-price data comprising 2,735 observations from 2015 to 2025 were collected and normalized using Min-Max Scaling. The data were divided chronologically into 80% training and 20% testing sets. A hybrid CNN–LSTM model was trained using the Adam optimizer with a learning rate of 0.0001, dropout of 0.2, a timestep of 30, and batch sizes of 16, 32, and 64. Model performance was evaluated using Root Mean Square Error (RMSE). Results and Discussion: The batch size of 16 achieved the best performance, producing the lowest validation RMSE of 0.0929 and an RMSE of 11.518535% after denormalization, outperforming batch sizes of 32 and 64. The model also followed actual gold-price trends more closely, while the inclusion of Dense and Dropout layers improved generalization. Conclusion: The CNN–LSTM hybrid model, particularly with a batch size of 16, provides an effective approach for forecasting volatile gold prices by integrating local pattern extraction with long-term temporal modeling.
Resin Code Classification on Plastic Packaging Using Few-Shot Learning Alyani Noor Septalia; Anindita Septiarini; Masna Wati
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.406

Abstract

Introduction: Accurate sorting of plastic waste using Resin Identification Codes (RICs) is essential for improving recycling quality. However, conventional deep learning approaches generally require large labeled datasets, which are difficult and costly to collect for small RIC symbols on plastic packaging. Method: This study employed a Few-Shot Learning approach based on Prototypical Networks using a 7-way 5-shot episodic training configuration. A self-collected dataset of 350 images, comprising 50 images for each of seven RIC categories—PETE, HDPE, PVC, LDPE, PP, PS, and OTHER—was used. Three backbone architectures, ConvNet4, ResNet-18, and EfficientNet-B2, were compared. An ablation study evaluated support-set augmentation, followed by supervised fine-tuning of the selected model. Results and Discussion: EfficientNet-B2 achieved the highest episodic accuracy of 93.36%, outperforming ResNet-18 at 88.71% and ConvNet4 at 54.14%. EfficientNet-B2 with light augmentation attained 85.71% accuracy on the fixed 42-image test set, with perfect recall for PVC, LDPE, and PP. Most errors involved visually similar HDPE and PP symbols. Fine-tuning corrected five of six misclassifications, increasing test accuracy to 90.48% and the F1-score from 0.857 to 0.903. Conclusion: Prototypical Networks with an EfficientNet-B2 backbone and cosine distance provide an effective solution for RIC classification under limited-data conditions and offer a practical foundation for automated plastic-waste sorting systems.
Comparing ECLAT and Decision Tree for Drug Therapy Recommendation Rules on Multi Label Clinical Data Muh Rayhan Fahreza Rayhan; Harlinda; Herdianti Darwis; Roesman Ridwan Raja
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.410

Abstract

Introduction: Accurate sorting of plastic waste using Resin Identification Codes (RICs) is essential for improving recycling quality. However, conventional deep learning approaches generally require large labeled datasets, which are difficult and costly to collect for small RIC symbols on plastic packaging. Method: This study employed a Few-Shot Learning approach based on Prototypical Networks using a 7-way 5-shot episodic training configuration. A self-collected dataset of 350 images covering seven RIC categories was used, and three backbone architectures, ConvNet4, ResNet-18, and EfficientNet-B2, were compared. An ablation study evaluated support-set augmentation, followed by supervised fine-tuning of the selected model. Results and Discussion: EfficientNet-B2 achieved the highest episodic accuracy of 93.36%, outperforming ResNet-18 at 88.71% and ConvNet4 at 54.14%. EfficientNet-B2 with light augmentation achieved 85.71% accuracy on the fixed 42-image test set. Most errors occurred between visually similar HDPE and PP symbols. Fine-tuning corrected five of six misclassifications, increasing test accuracy to 90.48% and the F1-score from 0.857 to 0.903. Conclusion: Prototypical Networks with an EfficientNet-B2 backbone and cosine distance provide an effective approach for RIC classification under limited-data conditions and offer a practical foundation for automated plastic-waste sorting systems.
An LLM-Based AI Task Agent for Academic Task Management with n8n and Telegram Nabila Widiyanti; Tasrif Hasanuddin; Huzain Azis
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.412

Abstract

Introduction: Managing multiple academic tasks with overlapping deadlines remains challenging for university students, while conventional task-management applications still require substantial manual organization and prioritization. This study develops an LLM-based AI Task Agent that enables conversational academic task management through Telegram and workflow automation. Method: The proposed system integrates Telegram as the interaction interface, n8n for workflow orchestration, an LLM-based AI agent for natural-language interpretation and tool selection, Google Sheets for task-data operations, PostgreSQL for conversational memory, and scheduled workflows for automated reminders. The system supports Create, Read, Update, and Delete operations, contextual priority recommendations based on deadline, urgency, and lecturer strictness, and proactive reminders. Functional performance and response time were evaluated across the primary system functions. Results and Discussion: Create, Read, Update, and Delete operations achieved 100% functional accuracy, while priority recommendation and automated reminder functions achieved 95%. Recorded processing times ranged from 3.1 to 3.5 seconds, with an average of approximately 3.32 seconds. The results demonstrate that separating LLM-based interpretation from predefined external tool execution enables reliable conversational task management while maintaining controlled data operations. Conclusion: The proposed LLM-based AI Task Agent demonstrates the feasibility of integrating conversational interaction, executable task-management functions, contextual prioritization, memory, and proactive reminders within a unified Telegram-based academic workflow.
A Pilot Study on Machine Learning Models for Predicting Student Pass/Fail Outcomes Using LMS Activity Logs and Interactive Dashboard Analytics Pratiwi Rachmadi
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.415

Abstract

Introduction: Learning Management Systems (LMS) generate extensive behavioral data that can support early identification of students at risk of academic failure, yet such data are often underutilized for predictive and intervention purposes. Method: This exploratory pilot study developed a machine learning pipeline following the Knowledge Discovery in Databases (KDD) framework using LMS activity logs from 36 students enrolled in a Discrete Mathematics course. Student-level behavioral features were extracted and modeled using Logistic Regression, Random Forest, and XGBoost. Model performance was evaluated using repeated 5-fold cross-validation, while an interactive learning analytics dashboard was designed to translate predictions into risk alerts and actionable recommendations for lecturers. Results and Discussion: Random Forest achieved the best overall performance with 87.0% accuracy, 0.88 precision, 0.94 recall, a macro-averaged F1-score of 0.92, and an AUC of 0.89, outperforming XGBoost and Logistic Regression. Assignment regularity and average quiz scores were identified as the most influential predictors. Qualitative feedback from two lecturers indicated that integrating risk information within a familiar LMS interface could support more practical and timelier student intervention. However, the small single-course sample limits generalizability. Conclusion: The proposed approach demonstrates the feasibility of integrating machine learning prediction and learning analytics dashboards as an early-warning mechanism, while further validation using larger, multi-course, and multi-institutional datasets is required.
Interpreting Reactive Species Interdependencies in Plasma-Activated Saline: An Exploratory Multivariate Workflow Nilton Francelosi Azevedo Neto; Orisson Ponce Gomes; Lucas Pereira Piedade; Felipe Souza Miranda; Rodrigo Savio Pessoa
Indonesian Journal of Data and Science Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science
Publisher : yocto brain

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.56705/ijodas.v7i2.417

Abstract

Introduction: Interpreting Plasma-Activated Saline (PAS) is challenging because reactive oxygen, nitrogen, and chlorine species are strongly interdependent, making it difficult to isolate their individual contributions to physicochemical properties such as Oxidation-Reduction Potential (ORP). This study explores these relationships using a multivariate chemometric workflow. Method: A small experimental dataset comprising 10 PAS observations generated by a serial DBD-GAPJ reactor was analyzed using Exploratory Data Analysis and Multiple Linear Regression. ORP was modeled as a function of O3, H2O2, HClO, NO3−, and NO2−. Model interpretation was supported by Variance Inflation Factor analysis, standardized coefficients, Leave-One-Out Cross-Validation, and residual diagnostics. Results and Discussion: H2O2 showed the strongest bivariate correlation with ORP (r = 0.86). The regression model achieved R² = 0.84 and adjusted R² = 0.65, but LOOCV produced Q² = −0.33, indicating poor out-of-sample prediction. Strong multicollinearity among reactive species complicated coefficient interpretation, and none of the individual predictors reached conventional statistical significance. Standardized coefficients identified H2O2 as the strongest relative contributor, while O3 and HClO remained chemically plausible interdependent contributors. Conclusion: The proposed workflow is valuable for interpreting small, collinear chemical datasets, but the results should be regarded as hypothesis-generating rather than predictive, emphasizing the importance of standardization, cross-validation, and multicollinearity diagnostics

Page 18 of 20 | Total Record : 191

 16   17   18   19   20 

Filter by Year

2020 2026


Reset
Filter By Issues
All Issue Vol. 7 No. 2 (2026): Indonesian Journal of Data and Science Vol. 7 No. 1 (2026): Indonesian Journal of Data and Science Vol. 6 No. 3 (2025): Indonesian Journal of Data and Science Vol. 6 No. 2 (2025): Indonesian Journal of Data and Science Vol. 6 No. 1 (2025): Indonesian Journal of Data and Science Vol. 5 No. 3 (2024): Indonesian Journal of Data and Science Vol. 5 No. 2 (2024): Indonesian Journal of Data and Science Vol. 5 No. 1 (2024): Indonesian Journal of Data and Science Vol. 4 No. 3 (2023): Indonesian Journal of Data and Science Vol. 4 No. 2 (2023): Indonesian Journal of Data and Science Vol. 4 No. 1 (2023): Indonesian Journal of Data and Science Vol. 3 No. 3 (2022): Indonesian Journal of Data and Science Vol. 3 No. 2 (2022): Indonesian Journal of Data and Science Vol. 3 No. 1 (2022): Indonesian Journal of Data and Science Vol. 2 No. 3 (2021): Indonesian Journal of Data and Science Vol. 2 No. 2 (2021): Indonesian Journal of Data and Science Vol. 2 No. 1 (2021): Indonesian Journal of Data and Science Vol. 1 No. 3 (2020): Indonesian Journal of Data and Science Vol. 1 No. 2 (2020): Indonesian Journal of Data and Science Vol. 1 No. 1 (2020): Indonesian Journal of Data and Science More Issue