Komputasi: Jurnal Ilmiah Ilmu Komputer dan Matematika

Real-Time Detection of Huanglongbing (HLB) Disease in Citrus Leaves Using Enhanced YOLO V8 Algorithm

Fri, 30 Jan 2026 00:00:00 +0000

This study addresses the complex challenge of detecting Huanglongbing (HLB) disease in citrus leaves, which is known as one of the most lethal plant diseases with no known cure. The primary issue in HLB detection is the difficulty in identifying symptoms early and accurately, particularly in dynamic and uncontrolled field environments. Therefore, the main focus of this research is the development of a real-time detection approach using the YOLO V8 algorithm to more accurately detect and classify HLB symptoms in citrus leaf images. The objective of this study is to design a technique that can enhance the detection of HLB disease and compare its performance with the conventional YOLO V8 method. This research also aims to address the limitations of previous studies that used the Support Vector Machine (SVM) method, which only achieved an accuracy of 80%. To achieve this objective, the study utilizes a dataset consisting of 1200 citrus leaf images, representing various levels of severity, including mild, moderate, severe, and healthy leaves. The method employed in this research involves the use of the YOLO V8 algorithm to detect and classify HLB symptoms in citrus leaf images. This approach was tested through a series of experiments to measure accuracy, precision, recall, and computational efficiency. The experimental results consistently demonstrate that the developed approach outperforms the basic YOLO V8 and previous methods using SVM, with an improvement in HLB disease detection accuracy reaching 98%. This study provides critical insights into early detection of HLB disease, potentially serving as a powerful tool to support efforts in preventing the spread of this disease across citrus orchards. Additionally, this research opens opportunities for further development in real-time plant disease detection by integrating more advanced AI technologies and applying similar methods to other plant diseases. Future research can focus on developing more efficient and scalable algorithms for use in various field conditions, as well as exploring the integration of sensors and IoT technology for more comprehensive plant health monitoring.

Implementation of Convolutional Neural Network with VGG-16 Architecture in Digital Hiragana Handwriting Image Recognition

Fri, 30 Jan 2026 00:00:00 +0000

The number of Japanese language learners in Indonesia ranks second at 711,732 people. Hiragana is the first letter to be learned, especially at the beginner level and is usually learned before Katakana and Kanji. Some characters in Hiragana have similar main forms such as nu (ぬ) and me (め), ne (ね) and wa (わ), thus adding complexity to the recognition process. Like previous research that created a Hiragana pronunciation learning application and previous research that was an English writing learning application, allowing people to learn on their own, by applying CNN (Convolutional Neural Network) to recognize written characters, researchers were inspired to apply this in learning to write Hiragana letters. Therefore, researchers created a digital Hiragana handwriting recognition model using the VGG-16 CNN Architecture method so that the model created can later be used in a Hiragana learning application for writing. This study used a dataset in the form of digital Hiragana handwriting images totaling 1518 data with 33 data for each label (46 types of letters). The hyperparameters used in this study to train the model were 5 epochs, a batch size of 32, the Adam Optimizer, and a Learning rate of 0.001. Based on the test results with the aforementioned parameters, the Accuracy value was 98.55%, Precision was 98.91%, Recall was 98.55%, and the F1-Score was 98.51%.

Assessing Assessing Students’ Ethical Concerns in AI-Integrated Online Learning Systems: A Study in Batam City

Hendi Sama, Julianto, Surya Tjahyadi — Fri, 30 Jan 2026 00:00:00 +0000

The adoption of Artificial Intelligence (AI) in online education introduces ethical concerns related to accuracy, fairness, and accountability. This study examines students’ ethical concerns of AI-integrated learning systems, focusing on AI-generated materials, attendance monitoring, and chatbot interactions. A mixed-method approach combining qualitative and quantitative techniques was used, involving interviews with 48 students from three different educational levels in Batam City. Thematic analysis identified five dominant themes. These included AI as a useful yet unreliable learning assistant, concerns about fairness in AI-based monitoring, uncertainty regarding responsibility and accountability, the need for transparent institutional policies, and limited AI literacy leading to overreliance. The study reports its findings using percentage-based distributions to illustrate the prevalence of these concerns across educational levels. The results indicate that students’ acceptance of AI in online learning is closely tied to the presence of human oversight, transparent institutional policies, and clearly defined accountability mechanisms. The novelty of this study lies in its focus on a pre-adoption educational environment, where AI is not yet fully institutionalized. Unlike prior studies examining post-implementation contexts, this research captures students’ anticipatory ethical expectations, highlighting concerns often overlooked in retrospective evaluations. The study contributes by providing empirical evidence across educational levels, offering localized insights from a developing Indonesian city, and extending AI ethics research beyond technology-advanced settings. The findings emphasize the importance of human oversight, accountability mechanisms, and transparent institutional policies for ethically grounded AI governance in education.

Acquiring Knowledge from Data Analytics and Performance-Boosting on Multimedia Content

Hendi Sama, Jed Wan, Muhamad Dody Firmansyah — Fri, 30 Jan 2026 00:00:00 +0000

In gaining meaningful and actionable insights from complex and diverse multimedia content, many studies have applied data analytics approaches—particularly data mining and machine learning—to uncover patterns, relationships, and hidden knowledge. This systematic literature review synthesizes 26 studies conducted over the past decade on acquiring knowledge from multimedia content using data analytics and performance-boosting techniques. Across domains such as social media, education, healthcare, e-commerce, and public safety, most works integrate text–image or audio–video pairs and increasingly adopt attention-based architectures and transformer models with early fusion strategies. To ensure comparability, each study’s evidence is recorded by considering the reported performance improvement over the authors’ baseline using the same dataset and evaluation metric. The most frequently used metrics include Accuracy, the F1-score (a harmonic mean of Precision and Recall), Precision, Recall, and the Area Under the Receiver Operating Characteristic Curve (AUC), which provides a threshold-independent measure of classification quality. The most common challenges identified include modality integration and alignment, data noise and quality, limitations of datasets and benchmarks, and domain shift, with fewer studies reporting class imbalance, computational cost, and interpretability or privacy issues. At the same time, promising opportunities emerge in the development of standardized multimodal benchmarks, efficient transformer-based and hybrid fusion pipelines, integration of external knowledge, domain-robust learning, and privacy-preserving explainable multimodal artificial intelligence. Overall, this review contributes a consolidated map of modalities, methods, and metrics, a performance-gain versus baseline table for quick comparability, a quantified challenge landscape, and a practical roadmap for guiding future research in multimodal sentiment analysis and related fields.

Design and Development of Emergency Mobile Application Using Design Thinking and Agile Scrum: A Case Study of Batam City

Li Cen, Martius Lim, Ricky Roy Nardson, Herman — Fri, 30 Jan 2026 00:00:00 +0000

Emergency response in Batam City, an area with high urban mobility and dynamics, is often hampered
by slow reporting mechanisms. Residents must contact different numbers for each service in different
areas, such as the fire department, ambulance, or police, causing confusion and wasting valuable time,
especially in panic situations. The absence of an integrated system specifically creates significant
inefficiencies in emergency service response. This study aims to design and develop an integrated mobile
application that serves as a tool to accelerate the reporting and handling of emergency conditions and
public utility services in Batam City. The methodology used combines the Design Thinking approach,
which ensures that solutions are designed based on the real needs and experiences of users, with the Agile
Scrum method, which allows for a flexible and iterative development process. This research successfully
produced a functional application that was then comprehensively tested. Functionality testing using the
Usability Testing Method showed that all core features worked well, were valid, and performed as
expected. To measure usability, a System Usability Scale (SUS) test was conducted, resulting in an
exceptional score of 88.82. This score places the application in the “Excellent” category, with an
“Acceptable” user acceptance rating, indicating a highly intuitive and easy-to-use interface. The main
conclusion is that the developed application has proven to be highly viable, functional, and well-received
by users. This application shows significant potential as an effective tool for improving the speed and
efficiency of emergency service responses, and could serve as a model technological solution for public
service integration in the City of Batam.

Sentiment Analysis of Electric Vehicles on Social Media Using Bidirectional Encoder Representations from Transformers (BERT) and Long Short-Term Memory (LSTM)

Muhammad Fadhillah Harahap, Yusma Yanti, Prihastuti Harsani — Fri, 30 Jan 2026 00:00:00 +0000

Electric vehicles (EVs) are widely recognized as an environmentally sustainable alternative capable of reducing greenhouse gas emissions; however, their adoption in Indonesia remains limited. Data from the Indonesian Ministry of Transportation, as recorded in the Type Approval Registration System (SRUT), indicate that approximately 195,084 Battery Electric Vehicles (BEVs) were registered nationwide by early 2024. This study investigates public sentiment toward electric vehicles using social media data from X, Instagram, and TikTok, while also comparing the effectiveness of two text classification approaches: Bidirectional Encoder Representations from Transformers (BERT) and Long Short-Term Memory (LSTM). A total of 5,172 Indonesian-language comments were collected through crawling and scraping techniques using electricvehicle-related keywords over the period January 2021 to January 2025. The comments were categorized into five sentiment classes: very positive, positive, neutral, negative, and very negative. The analytical process followed the Knowledge Discovery in Databases (KDD) framework, including data preprocessing, transformation, classification, and evaluation using a confusion matrix. The results indicate that IndoBERT substantially outperformed LSTM, achieving an accuracy of 91% compared to 36% for LSTM. Sentiment analysis reveals a dominance of negative and very negative opinions, primarily reflecting public concerns regarding cost, performance, and maintenance of electric vehicles. These findings offer important insights for policymakers and the automotive industry in designing targeted promotion strategies, improving public awareness, and strengthening supporting infrastructure. Future research is encouraged to explore data augmentation techniques to improve model performance, particularly for deep learning models such as LSTM, in order to better support evidence-based electric vehicle adoption policies.

Application of Linear Regression and Random Forest Algorithms in Predicting Human Development Index (HDI)

Mulyati Mulyati, Nur Aynun Siregar, Khairunnisa — Fri, 30 Jan 2026 00:00:00 +0000

The Human Development Index (HDI) is an important indicator for assessing the welfare and quality of life of the population in a region. The different growth of the HDI between regions indicates the need for accurate data-based analysis and prediction. One of them is a predictive analysis technique using the Linear Regression Algorithm and Random Forest. This study compares the two algorithms to predict the Human Development Index based on Expected Years of Schooling, Average Years of Schooling, Life Expectancy and adjusted Per Capita Income. The research stages include data collection, data pre-processing, data analysis and model evaluation. The results show that the use of the K-Fold Cross Validation method with a value of K = 5 produces a more optimal linear regression model compared to the Random Forest model. This is indicated by a higher coefficient of determination (R²) value and lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).

A Analysis and Modeling Simulation King Kuphi Cafe Queuing System With Customer Arrival Variations Using Python

Fri, 30 Jan 2026 00:00:00 +0000

Queueing system modeling and simulation is an effective approach for analyzing service performance in business environments with dynamic customer arrival rates, such as at King Kuphi Cafe. This study aims to model the queueing system at the cafe with various variations in customer arrival rates using the queueing theory approach and simulate it using the Python programming language. The models used are the M/M/1 and M/M/c queueing systems, which allow analysis of changes in waiting time, queue length, server utilization, and service level based on variations in arrival (λ) and service (μ) parameters. The simulation was run using Python packages such as NumPy and SimPy to represent the arrival and service processes realistically. The results of the study show that an increase in the rate of customer arrivals significantly affects system performance, particularly in terms of an increase in average waiting time and queue length. In addition, adding more servers has been proven to reduce queue congestion and improve overall service quality. These findings are expected to serve as a basis for King Kuphi Cafe managers in making strategic decisions regarding the number of baristas and operational optimization to achieve more efficient service.

An Empirical Study of Temporal Graph Neural Networks for Dynamic Node Forecasting

Ricky Maulana Fajri, Tasmi Tasmi, Ni Wayan Pricila Yuni Praditya — Fri, 30 Jan 2026 00:00:00 +0000

Temporal graph modeling has become increasingly important for understanding and forecasting the dynamics of complex systems that evolve over time. One of the central challenges in temporal graph learning lies in identifying graph neural network (GNN) architectures that can effectively capture both spatial dependencies and temporal dynamics. This study presents a comprehensive benchmarking analysis of widely used GNN architectures, namely Graph Convolution Network (GCN), GraphSAGE, Graph Attention Network (GAT), Chebyshev Networks (ChebNet), and Simplified Graph Convolution Network (SGC), each integrated with recurrent mechanisms for temporal modeling. The evaluation is conducted on the WikiMaths dataset, a large-scale temporal graph dataset representing user visits of mathematics-related Wikipedia articles. Experimental results demonstrate that the choice of graph convolution operator significantly impacts temporal forecasting performance, with GraphSAGE and ChebNet consistently exhibiting superior performance compared to other architectures. This work provides empirical insights into the strengths and limitations of established temporal GNN models, contributing to a clearer understanding of their applicability in dynamic graph forecasting tasks.

Design and Implementation of a Teaching Assistant Information System Using Laravel Filament and Extreme Programming

Nasrul, Edi Wibowo, David Wahyu Wismanindra — Fri, 30 Jan 2026 00:00:00 +0000

The management of teaching assistants at Sekolah Tinggi Teknologi Terpadu Nurul Fikri (STT-NF) has traditionally been conducted using manual processes, resulting in time inefficiencies, data inaccuracies, and limited integration of information. To address these challenges, a web-based Teaching Assistant Information System was developed using the Laravel Framework with Filament as the administrative interface and PostgreSQL as the database management system. The system is designed to streamline teaching assistant recruitment, class and assistant scheduling, and honorarium calculation in a structured and efficient manner. The development process applies the Extreme Programming (XP) methodology, which emphasizes iterative development, intensive user involvement, and adaptability to changing requirements. The results indicate that the proposed system improves efficiency and data accuracy while supporting administrative and academic staff in monitoring activities and making informed decisions.

A Metaheuristic Hybrid Approach for University Timetabling- Genetic Algorithm and Simulated Annealing

Septian Cahyadi, Thesya Mercella — Fri, 30 Jan 2026 00:00:00 +0000

This study addresses the recurrent course scheduling problem in universities. The problem involves constructing an optimal timetable by allocating courses, lecturers, and student groups to rooms and time slots while satisfying mandatory hard constraints and improving quality through soft constraints. Given the scale—nine study programs, 148 courses, 123 classrooms, 82 class groups, and 147 active lecturers—the problem exhibits combinatorial complexity. We propose a hybrid metaheuristic that integrates Genetic Algorithm (GA) and Simulated Annealing (SA) to balance global exploration and local exploitation. GA is selected for its robust exploration of large solution spaces and its proven applicability to university timetabling, while SA offers principled local refinement guided by an annealing schedule to reduce constraint violations. Prior work indicates that GA–SA hybrids can improve convergence and reduce computation time relative to standalone GA. We formalize the constraints, define a fitness function that prioritizes feasibility, and design neighbourhood operators tailored to timetabling moves. The proposed approach aims to deliver a robust timetable that satisfies institutional requirements and enhances operational efficiency.

Floating Solar Power Plant Planning Design for Public Facility Needs in Situ Tunggilis

Fri, 30 Jan 2026 00:00:00 +0000

Situ Tunggilis is a regional government asset managed by the Public Works Department of West Java Province and is part of an optimization plan by the Central Government through the Ministry of Public Works. One way to utilize the Situ Tunggilis area is by developing it as a tourist destination. To support this, adequate infrastructure such as public street lighting and reliable internet facilities is essential. The implementation of a Floating Solar Power Plant offers a sustainable solution to meet these energy needs, especially considering the limited available land and the demand for operational cost efficiency. This study focuses on designing a floating solar power plant system to supply independent energy for lighting and internet services in the tourist area. The approach involves technical calculations based on field measurements and secondary data, combined with system performance simulations. The simulation results demonstrate that the floating solar power plants system can meet the electricity demand for lighting and internet services in the pedestrian area, achieving a performance ratio of 71% and utilizing approximately 75% of the available solar energy. Moreover, the planned Wi-Fi network meets the required internet coverage radius for the tourist area.