cover
Article Per Year (5 Year)
All
Home Page
OAI Link
Editorial Team
Contact
Reviewer
Google Scholar
Contact Name
Dahlan Abdullah
Contact Email
dahlan@unimal.ac.id
Phone
+62811672332
Journal Mail Official
ijestyjournal@gmail.com
Editorial Address
Jl. Tgk. Chik Ditiro, Lancang Garam, Lhokseumawe, Aceh - Indonesia, 24351
Location
Kota lhokseumawe,
Aceh
INDONESIA
International Journal of Engineering, Science and Information Technology
Published by Universitas Malikussaleh
ISSN : -     EISSN : 27752674     DOI : -
Core Subject : Science, Education, Social, Engineering,
Astronomy Biochemistry, Genetics & Molecular Biology Chemical Engineering, Chemistry & Bioengineering Chemistry Civil Engineering, Building, Construction & Architecture Computer Science & IT Control & Systems Engineering Decision Sciences, Operations Research & Management Earth & Planetary Sciences Education Electrical & Electronics Engineering Energy Engineering Industrial & Manufacturing Engineering Library & Information Science Materials Science & Nanotechnology Mathematics Mechanical Engineering Physics Social Sciences Transportation
The journal covers all aspects of applied engineering, applied Science and information technology, that is: Engineering: Energy Mechanical Engineering Computing and Artificial Intelligence Applied Biosciences and Bioengineering Environmental and Sustainable Science and Technology Quantum Science and Technology Applied Physics Earth Sciences and Geography Civil Engineering Electrical, Electronics and Communications Engineering Robotics and Automation Marine Engineering Aerospace Science and Engineering Architecture Chemical & Process Structural, Geological & Mining Engineering Industrial Mechanical & Materials Science: Bioscience & Biotechnology Chemistry Food Technology Applied Biosciences and Bioengineering Environmental Health Science Mathematics Statistics Applied Physics Biology Pharmaceutical Science Information Technology: Artificial Intelligence Computer Science Computer Network Data Mining Web Language Programming E-Learning & Multimedia Information System Internet & Mobile Computing Database Data Warehouse Big Data Machine Learning Operating System Algorithm Computer Architecture Computer Security Embedded system Coud Computing Internet of Thing Robotics Computer Hardware Information System Geographical Information System Virtual Reality, Augmented Reality Multimedia Computer Vision Computer Graphics Pattern & Speech Recognition Image processing ICT interaction with society, ICT application in social science, ICT as a social research tool, ICT in education
Arjuna Subject : Ilmu Komputer (semua kategori) - Ilmu Komputer (Lain-Lain)
Articles 775 Documents
 72   73   74   75   76 
Confidence-Gated Escalation in Production Multi-Agent Systems: Multi-Signal Uncertainty Scoring, Composable Threshold Policies, and Regulatory-Aligned Audit Logging Ashutosh Rana
International Journal of Engineering, Science and Information Technology Vol 6, No 1 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i1.2029

Abstract

Large language model confidence is an unreliable signal when used in isolation, particularly in agentic workflows where tool invocations may produce consequential or irreversible effects. This study develops a confidence-gated escalation framework for production multi-agent systems that integrates complementary uncertainty signals with context-aware escalation policies. The framework combines normalized log-probability, verbalized uncertainty detection, and a tool-call risk taxonomy into a weighted composite confidence score. A composable policy architecture subsequently maps confidence and risk conditions to four escalation mechanisms: human-in-the-loop review, model upgrading, tool restriction, and compliance logging. The compliance logging mechanism generates structured audit records containing decision-relevant information to support traceability and regulatory record-keeping. Framework adapters for LangChain, CrewAI, AutoGen, and Google ADK are incorporated to facilitate deployment within existing agentic pipelines without requiring substantial architectural modification. Validation across three anonymized production deployments demonstrates that multi-signal confidence scoring reduces false escalations by 23% compared with single-signal log-probability scoring while maintaining deployment-specific human review coverage. The evaluation further shows that explicit tool-risk assessment enables escalation decisions to account for the potential consequences of agent actions rather than relying solely on model-reported confidence. The results indicate that combining heterogeneous uncertainty signals with contextual risk information provides a more consistent basis for escalation decisions than empirical threshold calibration alone. The framework offers a lightweight, composable mechanism to improve operational reliability, human oversight, and auditability in production multi-agent systems while preserving flexibility across agent architectures, tool environments, and organizational compliance requirements
Benchmarking the Learnability of Aspect-Based Sentiment Analysis on Indonesian Protest Tweets Putri Rizqiyah; Ahmad Ridwan Fauzi; Adelia Chitra Sazkia
International Journal of Engineering, Science and Information Technology Vol 6, No 4 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i4.2006

Abstract

Aspect-based sentiment analysis (ABSA) on Indonesian social media has steadily improved in accuracy, yet the field still lacks evidence on which ABSA subtasks are learnable on imbalanced, event-driven text and how much of a reported score reflects the protocol rather than the model. This study benchmarks relevance, aspect, and aspect-level polarity classification on 7,131 tweets from the August 2025 Indonesian demonstrations, against a documented gold standard of 6,075 relevant tweets. Majority-class prediction, four lexical models, a fine-tuned IndoBERT, and zero-shot GPT-4o-mini are compared under identical five-fold stratified cross-validation, with macro-F1 as the primary metric, with 95% paired-bootstrap and McNemar tests on identical folds, and with a repeated benchmark under near-duplicate-grouped and day-held-out splits. Headline accuracy is misleading for relevance: models reach 85-89% against an 85.19% majority baseline, yet lexical models detect only one in ten off-topic tweets (recall 0.099) and the fine-tuned encoder no more than half (0.495). Aspect classification is the most learnable subtask, at 0.772 macro-F1 for the fine-tuned encoder, and zero-shot GPT-4o-mini reaches 0.808 on the 78.9% of records it chose to label, against 0.711 when its omissions are counted as errors. Positive polarity is the bottleneck for every family: lexical models reach 0.306 F1 with 0.190 recall, and although contextual and zero-shot models recover positive recall to roughly 0.45, F1 stays in the 0.51-0.57 range on the records they label. Sentiment toward the people aspect is the hardest cell at 57.5% accuracy, against 86.6% for the government. The zero-shot comparator labels 84.2% of the relevant tweets for polarity, and its errors are systematic: 68.2% over-predict the negative class, conflating event valence with evaluative polarity. The study contributes a task-level, coverage-aware benchmark for Indonesian ABSA under class imbalance
Comparative Evaluation of Visual Behavioral Metrics to Identify Dominant Indicators for Reliable Real-Time Fatigue Detection I Putu Agus Eka Darma Udayana; Ni Wayan Sumartini Saraswati; Ni Made Sintya Noviana Utami
International Journal of Engineering, Science and Information Technology Vol 5, No 4 (2025)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v5i4.2046

Abstract

This study evaluates and compares three widely used behavioral visual indicators for real-time fatigue detection to determine the most reliable decision metric for continuous monitoring applications. The indicators examined consist of the Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), and Percentage of Eye Closure (PERCLOS), which represent micro-sleep events, yawning behavior, and cumulative eyelid closure, respectively. The experimental dataset includes 300 open–closed eye images, 300 yawning–non-yawning mouth images, and 20 five-minute video sequences to measure accumulated eye-closure over time. EAR obtained an accuracy of 89%, precision of 0.91, recall of 0.86, and an AUC of 0.93, showing high sensitivity to micro-sleep events lasting under two seconds. However, its performance decreased by up to 14% when the head pose shifted from a frontal orientation, indicating geometric dependency. MAR produced an average accuracy of 90%, precision of 0.87, recall of 0.84, and an AUC of 0.88, demonstrating strong responsiveness to yawning, with peak accuracy reaching 98%. Nevertheless, its stability dropped by 18% under tilted head orientation and lip movement variations, making it less reliable as a standalone metric for long-term monitoring. In contrast, PERCLOS achieved perfect performance with 100% accuracy, precision of 1.00, recall of 1.00, and an AUC of 1.00. Its robustness is attributed to cumulative eye-closure evaluation rather than geometric landmark measurement, allowing detection to remain unaffected by head pose variation. Statistical significance testing further revealed that PERCLOS outperformed both EAR and MAR (p 0.05), confirming its suitability as the dominant decision indicator. These findings establish PERCLOS as the most reliable metric for real-time visual fatigue assessment, while EAR and MAR serve as complementary indicators for detecting early fatigue symptoms such as micro-sleep and yawning behavior
Generalizing Operational Design Domain Frameworks from Autonomous Vehicle Safety Engineering to General-Purpose Physical Artificial Intelligence Systems Vraj Patel
International Journal of Engineering, Science and Information Technology Vol 6, No 1 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i1.2001

Abstract

The Operational Design Domain (ODD) concept provides a structured framework for defining the operational boundaries within which an automated system is designed to function safely. Although ODD has become an established concept in automated driving and is increasingly reflected in regulatory requirements, no equivalent cross-platform taxonomy has been established for general-purpose Physical Artificial Intelligence (Physical AI), including humanoid robots, mobile manipulators, and other physically embodied intelligent systems. This study examines the applicability of the ODD concept to Physical AI. It identifies the dimensions required to develop a unified taxonomy for describing operational boundaries across heterogeneous physical AI platforms. The study adopts a conceptual comparative approach by analyzing the principal dimensions used to characterize automated-vehicle ODDs and comparing them with the operational characteristics and domain-related provisions in standards and frameworks for industrial and service robots. The analysis indicates that several ODD dimensions, including environmental conditions, operational areas, dynamic agents, infrastructure, and system capabilities, can be generalized to Physical AI. However, the broader operational variability of Physical AI requires additional dimensions addressing human–robot interaction, task context, workspace configuration, manipulation requirements, and platform mobility. Existing robotic standards contribute complementary concepts for safety, environmental constraints, human interaction, and operational conditions, but they do not provide a unified taxonomy applicable across different physical AI platforms. The findings support a layered taxonomy of the Physical AI domain that integrates environmental, infrastructural, task, interaction, mobility, and system-capability dimensions. Such a taxonomy can establish a common basis for defining, communicating, and evaluating operational boundaries while accommodating differences in platform configuration, intended tasks, and operating environments
Evaluating Multivariate Zero-Shot Chronos-2 for Probabilistic Forecasting of Selected Indonesian Banking Stock Return Hozairi Hozairi; Anwari Anwari; Syariful Alim
International Journal of Engineering, Science and Information Technology Vol 6, No 4 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i4.1944

Abstract

Daily stock return forecasting is challenging because of low signal-to-noise ratios and changing market dynamics. Time-series foundation models such as Chronos-2 offer a zero-shot forecasting approach, but their effectiveness in emerging markets such as Indonesia remains insufficiently established. This study evaluates Chronos-2 for point and probabilistic forecasting of five LQ45 banking stocks (BBCA, BBNI, BBRI, BBTN, and BMRI). Daily price data from 2016–2026 were transformed into log returns and evaluated using a monthly rolling-origin design. Four configurations were examined: univariate, multivariate, multivariate with calendar covariates, and multivariate with market proxies. Point forecasts were evaluated using root mean squared error (RMSE) and directional accuracy, while probabilistic forecasts were assessed using pinball loss and 90% prediction interval coverage. In the ablation experiment, the multivariate configuration (C2-MV) achieved the lowest aggregate RMSE among the evaluated Chronos-2 configurations. In the multihorizon evaluation, C2-MV achieved lower aggregate RMSE than the random walk at horizons of 1 and 5 sessions (2.6327% vs. 2.6980% and 2.1477% vs. 2.1553%), but performed slightly worse at horizon 20 (2.0324% vs. 2.0305%). Directional accuracy ranged from 46% to 49%. However, the Diebold–Mariano test with Holm correction found no statistically significant predictive difference between C2-MV and the random walk, either in aggregate or consistently across individual stocks (p 0.05). Aggregate 90% prediction interval coverage ranged from 76.74% to 87.49%, indicating undercoverage. Overall, zero-shot multivariate Chronos-2 provides competitive but limited predictive performance, with benefits that are small, horizon-dependent, and not statistically significant relative to simple benchmarks. The findings should be interpreted cautiously because the sample was selected ex post and is limited to five Indonesian banking stocks
Autonomous Test Case Generation from API Specifications Using AI: A Hybrid LLM and Reinforcement Learning Framework Indra Kumar Mondal
International Journal of Engineering, Science and Information Technology Vol 6, No 3 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i3.2035

Abstract

The selection of network fabrics for large-scale artificial intelligence (AI) training has traditionally favored InfiniBand because of its high-performance computing heritage and tightly integrated communication capabilities. This study evaluates whether Ethernet-based fabrics, particularly Remote Direct Memory Access over Converged Ethernet version 2 (RoCEv2) and the emerging Ultra Ethernet architecture, can provide a credible alternative for large-scale AI training. The study employs a structured literature-based comparative analysis rather than an original hardware benchmark. The evaluation considers five dimensions: latency and tail latency behavior, scalability beyond 10,000 graphics processing units (GPUs), congestion control maturity, ecosystem and vendor openness, and total cost of ownership. Evidence from production deployment reports, congestion-control studies, and comparative analyses of Ethernet, Ultra Ethernet, and InfiniBand is synthesized to identify performance characteristics, architectural trade-offs, and infrastructure implications. The analysis indicates that modern lossless Ethernet fabrics have substantially reduced the performance gap for collective-communication workloads encountered in large-scale distributed AI training. RoCEv2-based architectures can achieve competitive communication performance when congestion management, quality-of-service mechanisms, and network configuration are carefully engineered. InfiniBand retains advantages in highly coupled and latency-sensitive training environments where predictable communication behavior and mature end-to-end optimization are prioritized. Ethernet-based approaches, however, provide greater vendor diversity, broader interoperability, and potentially more flexible cost structures. The emergence and standardization of Ultra Ethernet further extend Ethernet's suitability for AI-oriented networking by targeting the communication patterns and congestion characteristics of large-scale accelerator clusters. The findings indicate that network fabric selection should be based on workload characteristics, cluster scale, performance requirements, and infrastructure priorities, rather than on a universal preference for a single technology
GIS-Based Spatial Analytics for Poverty Mapping and Socio-Economic Vulnerability in Aceh Province, Indonesia Munawir Munawir; Susmanto Susmanto; Taufik Hidayat; Zulfan Zulfan
International Journal of Engineering, Science and Information Technology Vol 6, No 4 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i4.1914

Abstract

Poverty remains a persistent development challenge in Aceh Province, characterized by considerable regional disparities and multidimensional socio-economic vulnerability. This study analyzes poverty trends, maps poverty distribution, identifies priority high-poverty areas, and examines socio-economic vulnerability across 23 regencies and cities in Aceh using 2025 data. A descriptive quantitative approach was applied using secondary data from Statistics Indonesia (BPS). The analysis comprised descriptive statistics, Geographic Information System–based thematic mapping, poverty-level classification, urban–rural comparison, and evaluation of the poverty line, Poverty Gap Index (P1), and Poverty Severity Index (P2). The results show that Aceh’s poverty rate declined from 12.33% in March 2025 to 12.22% in September 2025, while the number of poor people decreased from 704.69 thousand to 703.33 thousand. However, substantial spatial disparities persisted, with poverty rates ranging from 5.45% in Banda Aceh City to 17.07% in Aceh Singkil Regency. Eight regencies—Aceh Singkil, Gayo Lues, Pidie, Bener Meriah, Pidie Jaya, Simeulue, West Aceh, and Nagan Raya—were classified as priority high-poverty areas. Rural areas exhibited greater vulnerability than urban areas due to limited infrastructure, market access, employment diversification, and dependence on the primary sector. The poverty line increased from IDR 676,247 to IDR 715,103 per capita per month, with food accounting for 76.56% of its composition. Despite the declining poverty rate, P1 increased from 1.836 to 2.055 and P2 from 0.420 to 0.523, indicating deeper poverty and greater inequality among poor households. These findings highlight the need for place-based poverty alleviation policies focused on rural economic development, infrastructure improvement, food-price stabilization, and targeted protection for chronically poor households
The Concord Framework: Event-Driven Multi-Cloud AI Orchestration for Cross-Partner Clinical Interoperability Varun Reddy Beem
International Journal of Engineering, Science and Information Technology Vol 6, No 2 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i2.2023

Abstract

Patient record duplication and manual reconciliation workflows between payer and provider organizations represent a durable failure mode in healthcare data infrastructure. Single-cloud interoperability solutions address this problem only partially; they introduce standardized data exchange, but they cannot sustain AI inference continuity when a cloud provider experiences a regional outage. The Concord Framework takes a different approach: it treats multi-cloud distribution not as a redundancy add-on, but as the core architectural pattern from which both interoperability and resilience derive. Deployed across eight partner organizations in a six-month production pilot, the framework uses a cross-cloud event relay between Azure Event Grid and Amazon EventBridge as its primary integration layer, a two-model consensus architecture pairing Azure OpenAI GPT-4o with Amazon Bedrock Claude 3.5 Sonnet for high-stakes entity classification decisions, and an autonomous failover controller with a recovery time objective under four hours and a recovery point objective under fifteen minutes. Pilot outcomes include a 75% reduction in duplicate patient records, an 83% reduction in care-coordination preparation time (from 4 hours to 20 minutes), a 55% improvement in care-gap identification accuracy, and zero unclassified protected health information incidents throughout the pilot period. Partner onboarding duration was reduced from eight weeks to under two weeks. During a documented Azure OpenAI Service regional outage, the framework maintained a zero reconciliation backlog without operator intervention, and both failover events met the defined RTO and RPO targets. These results demonstrate that event-driven multi-cloud orchestration is not only feasible at production scale but also produces measurably better reliability outcomes than single-cloud approaches
From Navigation to Intent: A Capability-Exposure Framework for Agent-Ready Enterprise Portals Sandeep Gusain
International Journal of Engineering, Science and Information Technology Vol 6, No 1 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i1.2058

Abstract

Enterprise portals have traditionally served as navigational interfaces for service discovery, deterministic actions, and permission-scoped access. The emergence of agentic artificial intelligence introduces a different interaction model in which user intent can be interpreted, decomposed, and executed through multi-step actions across heterogeneous enterprise workflows and operational environments at scale. This study examines the architectural changes required when enterprise portals evolve from navigation-oriented interfaces into environments capable of hosting and governing agentic interactions. The study introduces the Capability-Exposure Framework (CEF), a four-dimensional architectural model comprising intent-oriented discovery, shared context management, capability-service exposure, and governed autonomy. The framework is examined through a qualitative case analysis of the Microsoft Azure Portal and its documented integration of agentic capabilities, followed by a structural comparison with agentic capabilities documented for the AWS Management Console and Google Cloud console. The analysis shows that agentic portal integration requires more explicit permission scoping, tenancy isolation, contextual controls, and auditability than conventional portal operation. It also identifies cross-platform convergence toward explicit confirmation mechanisms for agent-initiated state changes and toward service interfaces that expose machine-interpretable capabilities rather than data-oriented operations alone. The results indicate that effective agentic integration depends on coordinated changes across the portal, service, identity, context, and governance layers. The study does not evaluate quantitative performance metrics; instead, it contributes a structural vocabulary for analyzing portal modernization and a migration perspective for enterprise architects and platform teams. The findings establish agent integration as an architectural transformation involving service capabilities and governance mechanisms, rather than merely an additional user interface feature
Marginal Network Benefit-Aware Renewable DG Planning in Radial Distribution Systems Trieu Ngoc Ton
International Journal of Engineering, Science and Information Technology Vol 6, No 4 (2026)
Publisher : Malikussaleh University, Aceh, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.52088/ijesty.v6i4.1980

Abstract

High penetration of renewable distributed generation (DG) does not necessarily imply efficient network utilization, because additional installed capacity may provide progressively smaller improvements in distribution-system performance. This paper proposes a marginal network benefit-aware planning framework for renewable DG in radial distribution networks (DNs). The framework combines a Composite Network Benefit (CNB), Marginal Renewable Benefit Index (MRBI), Normalized Marginal Benefit (NMB), and Renewable Benefit Saturation Point (RBSP) to quantify both overall network improvement and the incremental technical value of additional DG capacity. A Benefit-Guided Artificial Bee Colony (BGABC) algorithm is employed to optimize DG locations and capacities subject to voltage, generation, and power-balance constraints. The proposed approach is evaluated on the 33-bus and 69-bus radial distribution systems over renewable penetration levels from 20% to 70% and compared with conventional ABC, PSO, DE, SA, and GWO under consistent simulation conditions. Twenty independent runs are also performed to assess statistical robustness. The results show that active power loss, voltage deviation, and grid-related CO? emissions continue to improve as DG penetration increases, whereas the marginal network benefit declines progressively. The RBSP is identified at 60% penetration for the 33-bus system, corresponding to 2.232 MW of DG, 88.7 kW active power loss, and an NMB of 0.188, and at 70% for the 69-bus system, corresponding to 2.660 MW of DG, 143.8 kW active power loss, and an NMB of 0.105. These results confirm that renewable-benefit saturation is feeder-dependent and that additional DG may remain technically feasible beyond the point where its incremental network value becomes limited. The proposed framework therefore provides a complementary planning criterion for identifying effective DG allocations and technically meaningful renewable penetration levels in radial DNs

Page 74 of 78 | Total Record : 775

 72   73   74   75   76 

Filter by Year

2021 2026


Reset
Filter By Issues
All Issue Vol 6, No 4 (2026) Vol 6, No 3 (2026) Vol 6, No 2 (2026) Vol 6, No 1 (2026) Vol 5, No 4 (2025) Vol 5, No 3 (2025) Vol 5, No 2 (2025) Vol 5, No 1 (2025) Vol 4, No 4 (2024) Vol 4, No 3 (2024) Vol 4, No 2 (2024) Vol 4, No 1 (2024) Vol 3, No 4 (2023) Vol 3, No 3 (2023) Vol 3, No 2 (2023) Vol 3, No 1 (2023) Vol 2, No 4 (2022) Vol 2, No 3 (2022) Vol 2, No 2 (2022) Vol 2, No 1 (2022) Vol 1, No 4 (2021) Vol 1, No 3 (2021) Vol 1, No 2 (2021) Vol 1, No 1 (2021) More Issue