Jurnal Teknik Informatika (JUTIF)
Jurnal Teknik Informatika (JUTIF) is an Indonesian national journal, publishes high-quality research papers in the broad field of Informatics, Information Systems and Computer Science, which encompasses software engineering, information system development, computer systems, computer network, algorithms and computation, and social impact of information and telecommunication technology. Jurnal Teknik Informatika (JUTIF) is published by Informatics Department, Universitas Jenderal Soedirman twice a year, in June and December. All submissions are double-blind reviewed by peer reviewers. All papers must be submitted in BAHASA INDONESIA. JUTIF has P-ISSN : 2723-3863 and E-ISSN : 2723-3871. The journal accepts scientific research articles, review articles, and final project reports from the following fields : Computer systems organization : Computer architecture, embedded system, real-time computing 1. Networks : Network architecture, network protocol, network components, network performance evaluation, network service 2. Security : Cryptography, security services, intrusion detection system, hardware security, network security, information security, application security 3. Software organization : Interpreter, Middleware, Virtual machine, Operating system, Software quality 4. Software notations and tools : Programming paradigm, Programming language, Domain-specific language, Modeling language, Software framework, Integrated development environment 5. Software development : Software development process, Requirements analysis, Software design, Software construction, Software deployment, Software maintenance, Programming team, Open-source model 6. Theory of computation : Model of computation, Computational complexity 7. Algorithms : Algorithm design, Analysis of algorithms 8. Mathematics of computing : Discrete mathematics, Mathematical software, Information theory 9. Information systems : Database management system, Information storage systems, Enterprise information system, Social information systems, Geographic information system, Decision support system, Process control system, Multimedia information system, Data mining, Digital library, Computing platform, Digital marketing, World Wide Web, Information retrieval Human-computer interaction, Interaction design, Social computing, Ubiquitous computing, Visualization, Accessibility 10. Concurrency : Concurrent computing, Parallel computing, Distributed computing 11. Artificial intelligence : Natural language processing, Knowledge representation and reasoning, Computer vision, Automated planning and scheduling, Search methodology, Control method, Philosophy of artificial intelligence, Distributed artificial intelligence 12. Machine learning : Supervised learning, Unsupervised learning, Reinforcement learning, Multi-task learning 13. Graphics : Animation, Rendering, Image manipulation, Graphics processing unit, Mixed reality, Virtual reality, Image compression, Solid modeling 14. Applied computing : E-commerce, Enterprise software, Electronic publishing, Cyberwarfare, Electronic voting, Video game, Word processing, Operations research, Educational technology, Document management.
Articles
1,314 Documents
<![CDATA[Comparative Analysis of TF-IDF, Word2Vec, and BERT with Logistic Regression, Random Forest, and XGBoost for Sarcasm Detection in News Headlines ]]>
<![CDATA[A'zham Albar Rasyid]]>;
<![CDATA[Pulung Nurtantio Andono]]>;
<![CDATA[Ruri Suko Basuki]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5803
<![CDATA[The detection of sarcasm in news headlines remains a significant challenge in Natural Language Processing due to the inherent ambiguity and limited contextual information in micro-texts. This research aims to evaluate the effectiveness of various text representation techniques and classification algorithms to identify the most optimal and computationally efficient model for sarcasm detection. The study examines three generations of text representation (TF-IDF, Word2Vec, and BERT) and classification algorithms, including Logistic Regression, Random Forest, XGBoost, and Stacking Ensemble, while comparing heavy and light preprocessing scenarios. The results reveal that light preprocessing which retains stopwords is more optimal for maintaining the semantic integrity of short texts. The combination of Logistic Regression and TF-IDF achieved the best performance with an accuracy of 0.8320 and an F1-Score of 0.8052. This model proved highly efficient with a training time of only 0.1397 seconds, significantly outperforming complex models like BERT and Stacking, which required up to 115.87 seconds without yielding a proportional increase in accuracy. These findings contribute to the field of NLP by demonstrating that for micro-text characteristics, frequency-based representations with minimal preprocessing offer a more effective and resource-efficient solution than complex contextual models, providing a benchmark for real-time sarcasm detection systems.]]>
<![CDATA[Optimizing Diabetes Complication Prediction Using Adaptive Boosting With Naive Bayes, Synthetic Minority Oversampling Technique, And Information Gain Feature Selection ]]>
<![CDATA[Farid Fitriyadi]]>;
<![CDATA[Erna Indriastiningsih]]>;
<![CDATA[Indriyati Indriyati]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5812
<![CDATA[Primary healthcare facilities often face significant technical challenges in the early detection of diabetes complications due to their heavy reliance on micro-datasets characterized by extreme class imbalances. Such data constraints typically make standard single-prediction models prone to overfitting and low sensitivity, which are critical issues in clinical decision-making. This study proposes a multi-level optimization framework to overcome these limitations by integrating Information Gain for dimensionality reduction, Synthetic Minority Over-sampling Technique (SMOTE) to handle class imbalance, and Adaptive Boosting (AdaBoost) with Categorical Naive Bayes as the base learner. To address the critical aspect of model validity on a highly limited dataset (n=20), this methodology employs a stratified cross-validation scheme to ensure reliable performance estimation and minimize prediction bias. The integration of Information Gain effectively optimized the model by reducing the predictor variables from seven to four optimal features, thereby decreasing computational complexity. Based on rigorous testing, the proposed ensemble architecture achieved a robust average accuracy of 87.00% and, more importantly, a high Recall (Sensitivity) of 90% in identifying positive complication cases. This research provides a significant scientific contribution to machine learning development by demonstrating that the systematic integration of SMOTE and AdaBoost offers a stable, precise, and computationally efficient solution for predictive modeling in data-constrained clinical environments. These findings suggest that the proposed model can serve as a reliable early warning system to improve patient outcomes in primary care settings.]]>
<![CDATA[Performance Evaluation of Fine-Tuned Whisper Models for Low-Resource Lampung Language Speech Recognition ]]>
<![CDATA[Aulia Indrawati]]>;
<![CDATA[Amalia Zahra Zahra]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5814
<![CDATA[The Lampung language possesses significant cultural and linguistic value, yet it remains digitally low-resource, presenting severe bottlenecks for Automatic Speech Recognition (ASR). This study addresses the technical challenges of low-resource ASR by performing a rigorous performance evaluation of fine-tuned Whisper models on a constrained local cross-dialect corpus. The dataset consists of 949 custom audio recordings (approximately 1.5 hours) capturing both Dialect A (Pesisir) and Dialect O (Abung) with manual ground-truth transcriptions. The engineering pipeline encompasses audio signal normalization, conversion to 16 kHz mono-channel, augmentation, and log-Mel spectrogram feature extraction. Three architectural scales are evaluated: Whisper-Tiny, Whisper-Base, and Whisper-Small. Evaluated via Word Error Rate (WER), Whisper-Small achieves the state-of-the-art performance with a test WER of 15.57% (validation WER 18.66%), significantly outperforming Whisper-Tiny (16.10%) and Whisper-Base (32.32%). Structural analysis reveals that Whisper-Small provides the optimal empirical balance between capacity representation and generalization stability under strict data scarcity. This research contributes a reproducible blueprint to the domain of computer science and speech processing, demonstrating that weak-supervision transformer architectures can effectively mitigate data sparsity in regional minority languages.]]>
<![CDATA[Bayesian-Augmented Rule-Based Recommender System Integrated with Customer Relationship Management to Enhance Congregant Retention in Religious Non-Profit Organizations ]]>
<![CDATA[Mychael Maoeretz Engel]]>;
<![CDATA[Yuwono Marta Dinata]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5839
<![CDATA[This study aims to address the lack of systematic engagement monitoring tools in religious non-profit organizations by developing a CRM-integrated recommender system that supports congregant retention. The research follows a two-phase approach: a systematic literature review (SLR) and a minimum viable product (MVP) implementation. The SLR analyzed 15 scientific articles (2020–2025), confirming that recommender systems in non-profit contexts are predominantly documented for educational, humanitarian, and healthcare sectors, with no prior work targeting religious organizations. Based on these findings, a web-based MVP was developed and deployed at GKI Citraland Surabaya. The system computes a quantitative Attendance Risk Score (RA) using a Bayesian Average formulation that combines individual absence history with a global prior mean, and presents at-risk congregants to administrators who can then initiate semi-manual WhatsApp reminders with one click. Black-box testing across 18 scenarios achieved 100% pass rate. Pilot validation confirmed a 95% reduction in monitoring time and detection of four previously unnoticed at-risk congregants. The primary contributions are: (1) empirical confirmation of a research gap in religious non-profit recommender systems; (2) a replicable Bayesian-augmented rule-based architecture for early warning engagement monitoring; and (3) field evidence of CRM-based retention analytics for community-based non-profit organizations.]]>
<![CDATA[Indonesian Sign Language Letter Recognition Using Mediapipe Hand Landmark Extraction And Random Forest Classification ]]>
<![CDATA[Barry Ceasar Octariadi]]>;
<![CDATA[Alda Cendekia Siregar]]>;
<![CDATA[Sucipto Sucipto]]>;
<![CDATA[Azrul Fathany]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5841
<![CDATA[This study aims to recognize letter patterns in Indonesian Sign Language (BISINDO) using the Random Forest algorithm. BISINDO represents 26 alphabet letters through distinct hand gestures. To automate the recognition process, this research utilizes MediaPipe for real-time feature extraction. MediaPipe detects and tracks hand movements using handedness to distinguish between left and right hands, as well as landmarks and world landmarks to capture key coordinate points that serve as input features for classification. The model is evaluated under various conditions, including differences in distance and lighting. Experimental results show that the model performs optimally at a distance of 30 cm. Under normal lighting conditions, the system achieves an accuracy of 88% by correctly recognizing 23 classes in the first test, and improves to 96% with 25 recognized classes in the second test, resulting in an average accuracy of 92%. However, under low-light conditions at the same distance, performance decreases, where only 9 out of 13 tested classes are correctly recognized, yielding an accuracy of 69%. When the distance increases to 60 cm, the model’s performance further declines, with 22 classes recognized at 84% accuracy in the first test and 20 classes at 76% accuracy in the second test. Further evaluation using a confusion matrix shows excellent classification performance, achieving 100% accuracy, recall, and precision. Overall, the integration of MediaPipe and Random Forest proves effective for BISINDO letter recognition and contributes to the development of assistive technologies that support inclusive communication through computer vision and machine learning approaches.]]>
<![CDATA[Adaptive Bare-Metal Data Center Load Distribution Using Dual-Policy Deep Reinforcement Learning with Hysteresis Switching and Safety Shielding ]]>
<![CDATA[Aay Ramdan]]>;
<![CDATA[Muhammad Faris Fathoni]]>;
<![CDATA[Setyorini Setyorini]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5844
<![CDATA[Load balancing in bare-metal server environments presents a unique challenge, as conventional static algorithms and single-agent Deep Reinforcement Learning (DRL) often fail to adapt securely to extreme traffic variations. Single DRL agents face a convergence dilemma: Advantage Actor-Critic (A2C) suffers from policy degradation under sudden stress, while Proximal Policy Optimization (PPO) lacks rapid responsiveness in low-traffic states. To address these limitations, this study proposes a Hybrid Adaptive Load Balancing (HALB) architecture that dynamically integrates A2C and PPO policies using hysteresis switching logic, fortified by a deterministic Safety Shield. The system was evaluated using an Nginx reverse proxy against standalone A2C and PPO baselines under Ramp-up, Sudden Spike, and High Traffic scenarios. Experimental results demonstrate that HALB provides noticeable improvements over single-policy agents, particularly in latency optimization and tail latency truncation. During extreme sudden traffic spikes, the HALB system constrained the 95th Percentile (P95) latency to 77.33 ms, compared to standalone A2C (85.00 ms) and PPO (85.23 ms). Under sustained high traffic conditions, HALB achieved an overall average latency of 23.64 ms while maintaining a load distribution fairness index (J > 0.99). Fortified by the Safety Shield that prevented node crashes at an 85% CPU threshold, the HALB architecture offers a balanced approach between intelligent responsiveness and operational stability. This research provides a resilient traffic management framework for rigid infrastructures, contributing to the development of safe adaptive control in non-elastic distributed systems.]]>
<![CDATA[Comparative Evaluation of Retrieval Augmented Generation Based Large Language Model with Formative Feedback and Traditional Instruction in Learning Fraction ]]>
<![CDATA[Ida Bagus Indrabudhi Kusuma]]>;
<![CDATA[Dade Nurjanah]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5847
<![CDATA[Learning fractions poses challenges for students, and while traditional one-on-one tutoring is effective, it faces human resource limitations. While generative artificial intelligence (AI) offers scalable personalized support, large language models (LLMs) risk generating hallucinations, leading students to misconceptions. To address these challenges, this study evaluates the effect of a Retrieval-Augmented Generation (RAG) based Intelligent Tutoring System (ITS) integrated with formative feedback on students' learning performance in fraction arithmetic, comparing it to traditional instruction. A quasi-experimental study was conducted with 56 7th grader students, divided into an experimental group (N=26) utilizing the ITS and a control group (N=30) receiving regular classroom instruction. The ITS employed GPT-4o and RAG grounded with fraction educational content to deliver personalized, step-by-step formative feedback based on the Mason and Bruning framework. Students’ learning performance was measured through pre and post-test and were analyzed using ANCOVA. The experimental group outperformed the control group with higher learning gain of 47.02% compared to the control group’s 31.54%. ANCOVA results confirmed the significant main effect (p = 0.002) suggesting that the RAG-based tutor improved the mathematical comprehension and reduced knowledge variance among students. These results underscore the grounded LLM integrated with formative feedback demonstrates potential to improve students learning performance. This study provides a comparative evaluation of RAG-based ITS integrated with formative feedback and traditional classroom instruction in improving student learning performance in fraction. Furthermore, this study advances AI-supported tutoring by introducing an LLM-based ITS that combines RAG-grounded generation, formative feedback, and LLM-based judgment for dynamic and adaptive student modelling.]]>
<![CDATA[A Hybrid Feature Fusion Framework Combining TF-IDF, N-Gram, and Lexicon Features for Tesla Review Sentiment Classification Using SVM ]]>
<![CDATA[Anggia Dasa Putri]]>;
<![CDATA[Rahmat Fauzi]]>;
<![CDATA[Sarjon Defit]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5850
<![CDATA[The Existing sentiment classification approaches that rely on single-feature representations often fail to capture the complexity of social media text, leading to suboptimal classification performance. This limitation is particularly evident in product-specific opinion mining, such as Tesla-related discussions on X (formerly Twitter), where textual expressions are highly diverse and context-dependent. Therefore, this study aims to develop a hybrid feature engineering approach that integrates multiple feature representations to improve sentiment classification performance. A dataset of 12,184 tweets was collected and automatically labeled into positive, neutral, and negative classes using a lexicon-based annotation method. The data were preprocessed through case folding, cleaning, tokenization, stopword removal, and stemming. Three types of features were extracted, including TF-IDF for statistical representation, N-gram for contextual information, and lexicon-based sentiment scores for semantic polarity. These features were combined into a unified hybrid feature vector and classified using a Support Vector Machine (SVM) model. The experimental results demonstrate that the proposed hybrid approach outperforms individual feature models. The hybrid model achieved an accuracy of 88.96% and an F1-score of 88.02%, with improvements of 6.82%, 5.69%, and 10.33% compared to TF-IDF, N-gram, and lexicon-only approaches, respectively. Additionally, the RBF kernel consistently outperformed the linear kernel, indicating the presence of non-linear patterns in sentiment data. These findings highlight the effectiveness of combining statistical, contextual, and semantic features in enhancing sentiment classification performance. This study contributes a practical and scalable hybrid feature framework for product-level sentiment analysis, which can be extended to other domains such as electric vehicles and consumer technology.]]>
<![CDATA[Hybrid Transformer-Graph Topic Modeling for Fine-Grained Analysis of Pluang App Reviews ]]>
<![CDATA[Zidni Ilmatun Nafiah]]>;
<![CDATA[Khothibul Umam]]>;
<![CDATA[Maya Rini Handayani]]>;
<![CDATA[Masy Ari Ulinuha]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5861
<![CDATA[The rapid growth of mobile apps has overwhelmed developers with a large number of unorganized user reviews, which are difficult to analyze efficiently because traditional methods like Latent Dirichlet Allocation (LDA) often fail to capture the deeper meaning behind user language. To address this issue, our study introduces a novel hybrid topic modeling approach specifically designed to decipher user feedback for the Pluang app by combining state-of-the-art transformer-based AI with graph modeling. In our framework, we use Sentence-BERT (SBERT) to capture the true meaning of text, construct a network graph to map how different comments are connected, and apply Louvain community detection to cluster similar discussions, while evaluating performance using standard metrics such as topic coherence, silhouette score, and topic diversity. After collecting 2,345 reviews from the Google Play Store and refining them into a clean sample of 2,295, our experiments revealed that while the traditional LDA model had a higher coherence score (0.5367), our hybrid model successfully uncovered 15 specific and highly detailed topics that LDA completely missed even with a slightly lower coherence score of 0.4342. Ultimately, by combining semantic embedding with graph clustering, this hybrid approach provides a much clearer and more nuanced understanding of users' true concerns, helping developers make better, data-driven decisions for future mobile app updates.]]>
<![CDATA[Classification of Community Aspiration Proposal Documents Using Large Language Model and Naïve Bayes Approach ]]>
<![CDATA[Muhammad Fariz Arya Brahmantya Putratama]]>;
<![CDATA[Surya Sumpeno]]>;
<![CDATA[Reza Fuad Rachmadi]]>
<![CDATA[ Jurnal Teknik Informatika (Jutif) Vol. 7 No. 5 (2026): JUTIF Volume 7, Number 5, October 2026 ]]>
Publisher : <![CDATA[Informatika, Universitas Jenderal Soedirman ]]>
Show Abstract
|
Download Original
|
Original Source
|
Check in Google Scholar
|
DOI: 10.52436/1.jutif.2026.7.5.5886
<![CDATA[Community proposals in the form of regional development proposals are often voluminous, creating administrative barriers to verification and alignment with dynamic development priorities. This study aims to implement and evaluate an automatic classification system to help verifiers quickly and accurately prioritize community proposals that are most relevant to the regional development policy directions. This study employs a comparative experimental design to evaluate the performance of a Large Language Model (LLM) against traditional machine learning models, namely Multinomial Naive Bayes (MNB) and Complementary Naive Bayes (CNB). PDF documents in the dataset were processed to extract text, and Optical Character Recognition (OCR) was used for scanned documents. Documents were sorted based on seven administrative criteria, then summarized to extract key points from those meeting the criteria, classified according to development priorities, and evaluated using Stratified 10-Fold Cross-Validation to ensure consistent and representative results. Experimental results show that Complement Naive Bayes (CNB) achieved the highest performance with an accuracy of 87.69% and an F1-score of 0.8760, followed by LLM with an accuracy of 87.72% and an F1-score of 0.8759, and MNB with an accuracy of 84.3% and an F1-score of 0.8430. Although CNB has a slight edge in statistical accuracy, LLM offers added value through its semantic evaluation capabilities, which can provide the logical reasoning behind each classification decision. This study provides empirical evidence that lightweight traditional algorithms (CNB) remain more robust and computationally efficient than zero-shot LLMs for localized administrative texts, offering a scalable solution for government digital transformation.]]>
