Bulletin of Computer Science Research

Analisis Faktor Penentu Profit Penjualan Mobil Menggunakan Algoritma Random Forest

Muhamad Fahrul Rozi — 2026-04-13

The automotive industry has significant changes in recent years that have directly affected vehicle sales profitability. The objective of this study is to analyze the factors influencing car sales profit using the USA Car Sales dataset for the 2018–2024 period. The approach employed is a quantitative method based on machine learning using the random forest algorithm, which was selected for its ability to handle complex data and identify important variables contributing to profit. The analysis was conducted through several stages, including data preprocessing, model training, performance evaluation, and result interpretation using feature importance techniques. These stages aim to obtain an accurate model while providing a comprehensive understanding of the influence of each variable on car sales profit. The results indicate that several factors have a significant impact on car sales profit, including car brand, year of sale, and the number of units purchased in a single transaction. Car brand reflects market preferences and consumer segmentation, while the year of sale represents market trends and changing conditions over time. In addition, the number of units sold per transaction plays an important role in increasing total profit. These findings provide strategic insights for automotive companies in formulating more effective, adaptive, and data-driven sales strategies.

Prediksi Kelulusan Mahasiswa Prodi Informatika dengan Algoritma Decision Tree (C4.5) dan Naïve Bayes

Steven Gerrard — 2026-04-13

The primary parameter for measuring higher education quality, which also has a crucial impact on the accreditation process, is the percentage of students graduating on time. However, the reality on the ground shows that many students face obstacles in completing their studies within the ideal timeframe. Therefore, a data-driven strategy is needed to project students' chances of graduation early. This research aims to compare the performance of the Decision Tree (C4.5) and Naïve Bayes algorithms in classifying the potential for on-time graduation. The data utilized included 161 entries from the Informatics Study Program, class of 2022, at the University of Muhammadiyah Sidoarjo. The attributes analyzed were divided into academic and non-academic factors, including gender, first-semester social studies grades (IPS), GPA, PKMU (Community Service Program) graduation score and status, BQ and Ibadah scores, and accumulated SKEK points. The research process went through several phases: preprocessing, class labeling, model development, and performance evaluation through a confusion matrix and 5-fold cross-validation. The test was validated by separating the training and test data into ratios of 70:30, 80:20, and 90:10. Based on the test results, the C4.5 algorithm achieved a peak accuracy of 100% across all ratio scenarios, with an average cross-validation accuracy of 96.88%. Meanwhile, Naïve Bayes achieved a maximum accuracy of 94.13% with an average cross-validation of 93.00%. These findings indicate that the C4.5 algorithm has superior performance on this specific dataset. The output of this predictive model is expected to serve as an objective basis for institutions in establishing proactive academic policies.

Evaluasi Aplikasi Pembelajaran Berbasis Web Menggunakan Generative Artificial Intelligence dengan Metode ROUGE

Rusmanto Rusmanto — 2026-04-13

This study aims to evaluate the functionality and answer quality of a web-based learning application that uses Generative Artificial Intelligence (GenAI) for the Pancasila and Civic Education (PPKN) course. The primary focus of this research lies in the system evaluation process, while the application development was carried out solely as a means of generating test data. The system was evaluated in two stages: functional testing using the black-box testing method and answer quality assessment using the Recall-Oriented Understudy for Gisting Evaluation (ROUGE) method. Black-box testing was conducted to ensure that all core system features operated according to specifications. The results of the black-box testing showed a 100% success rate across all test scenarios. Furthermore, answer quality evaluation was performed on 50 test data pairs consisting of GenAI-generated answers and reference texts (gold standards) prepared by PPKN lecturers using the ROUGE method. The evaluation results showed an average F1-score of 97% on the ROUGE-1, ROUGE-2, and ROUGE-L metrics. A total of 49 out of 50 answers were categorized as “Very Good” (? 0.75), while 1 answer was categorized as “Good.” These findings indicate that the application is capable of generating answers with a very high level of textual similarity to academic references. This study contributes to filling the gap in empirical evidence and provides a standardized evaluation benchmark for web-based GenAI applications in education, while also offering an evaluation approach that integrates system functional testing and ROUGE-based answer quality measurement. However, this evaluation is still limited to linguistic aspects based on n-grams and does not yet fully represent semantic depth.

Pengembangan Aplikasi E-Booking Konser K-Pop Berbasis QRIS dengan Pendekatan User-Centered Design untuk Optimalisasi Pengalaman dan Efisiensi Transaksi

Ozmar Azhari — 2026-04-13

The increase in the number of K-Pop concerts in Indonesia drives the need for a ticket e-booking system that is not only efficient but also capable of accommodating transaction characteristics with high demand levels and optimal integration of digital payment systems. The main issue with the current e-ticketing system lies in the platform's general nature (multi-event marketplace) and the suboptimal integration of user experience with digital payment systems like QRIS. This research aims to design and develop a mobile-based K-Pop concert e-booking application integrated with QRIS and empirically test its impact on transaction efficiency and user experience. The method used is Research and Development (R&D) with a System Development Life Cycle (SDLC) Waterfall model approach, complemented by a quasi-experimental design (posttest control group design). Testing was conducted on 60 respondents divided into experimental and control groups, with variables measured including transaction time, transaction success rate, usability using the System Usability Scale (SUS), and user satisfaction. The research results show that the K-Party application is capable of effectively and integratively supporting the e-booking process. Statistical analysis shows that the use of QRIS has a significant impact on transaction efficiency and user experience, as well as contributing to business performance improvements, including a 36.2% increase in revenue and a 5-10% increase in audience numbers. Thus, the developed system is not only technically feasible but also empirically proven to add value in the context of the digital entertainment industry.

Pengaruh Penggunaan User Centered Pada Perancangan UI/UX Pada Model Aplikasi Penjualan Es Teler Berbasis Website

Mutiara Nurikhlimah — 2026-04-22

Abstract The rapid development of information technology encourages businesses to adapt by utilizing digital platforms, one of which is through e-commerce websites. The Es Teler El website is developed as a web-based sales platform that requires optimal User Interface and User Experience (UI/UX) design to provide convenience, ease of use, and efficiency in transaction processes. This study aims to design the UI/UX of the Es Teler El website using the User-Centered Design (UCD) method and to evaluate the usability level of the resulting prototype. The UCD method is implemented through four main stages: Understand Context of Use, Specify User Requirements, Design Solution, and Evaluate Design Against Requirements. Data collection was conducted through interviews with 20 respondents to identify user needs. The results of the needs analysis were then implemented into wireframe, mockup, and high-fidelity prototype designs using Figma. Usability evaluation was carried out using the System Usability Scale (SUS) method involving the same respondents through predefined testing scenarios. The results show that the Es Teler El website prototype achieved an SUS score of 82.5, which falls into the “Excellent” category (Grade A). This indicates that the design meets usability aspects, including ease of use, efficiency, and user satisfaction. Therefore, the application of the UCD method is proven to be effective in producing UI/UX designs that align with user needs and improve the quality of user experience in web-based culinary e-commerce platforms.

Deteksi Pelanggaran Durasi Berhenti Kendaraan pada Area Yellow Box Junction Menggunakan Algoritma YOLOv8

Abdul Kholik — 2026-04-24

Traffic congestion at urban intersections in Indonesia, particularly in Palembang, is exacerbated by the frequent violation of Yellow Box Junction (YBJ) regulations. This study develops an automated detection system for vehicle stopping duration violations in YBJ areas using the YOLOv8 deep learning algorithm, specifically the yolov8n.pt model, optimized with a 3-frame skip technique to enhance computational efficiency. The system is designed to identify vehicles remaining within a predefined Region of Interest (ROI) for more than 5 seconds. Testing conducted at Simpang Angkatan 45 recorded 168 violations compared to 149 violations from manual observation. The primary contribution of this research lies in the development of an automated traffic law enforcement solution tailored to Indonesia's heterogeneous traffic conditions, as well as the implementation of computational optimization techniques that enable near real-time operation on mid-range hardware without significantly compromising detection accuracy . Although a 12.75% detection variance occurred due to ID switching and occlusion factors, this study provides a foundation for more accountable and scalable intelligent surveillance systems in the future.

Evaluasi Keamanan Website Direktori Akademik Menggunakan NIST SP 800-115

Fito Nardian — 2026-04-30

Evaluating the security of web-based academic information systems has become crucial as cyber threats in higher education environments increase. The track record of security incidents in information systems at UIN Sultan Syarif Kasim Riau has prompted an urgent need for preventative action; therefore, the website https://seminar-fst.uin-suska.ac.id, as an active academic service that stores sensitive data, requires a proactive evaluation. Testing used a black-box testing approach through four phases: planning, discovery, attack, and reporting. The results revealed a critical vulnerability in the form of SQL injection in URL parameters, which allows unauthorized database enumeration (MariaDB), thus threatening data confidentiality and integrity. Additionally, medium-level vulnerabilities were discovered, such as the use of an outdated JavaScript library (Moment.js 2.8.1) and misconfiguration of HTTP security headers, including the absence of a Content Security Policy (CSP) and an Anti-CSRF mechanism. Recommendations include prepared statements, strict input validation, updating dependencies, and strengthening security configurations.

Analisis Komentar Youtube Terhadap Kebijakan Bebas Impor Oleh Pemerintah Pusat Menggunakan Support Vector Machine

Ignasius Aditya Anggoro Putra — 2026-04-30

YouTube has become an important platform for expressing public opinion on government policies, including the free import policy. This study aims to analyze the sentiment of YouTube user comments regarding the free import policy using the Support Vector Machine (SVM) algorithm. The data were collected through web scraping using the YouTube Data API v3 from a Kompas.com video, resulting in 3,267 raw comments. The research stages include text preprocessing, feature extraction using Term Frequency–Inverse Document Frequency (TF-IDF), lexicon-based sentiment labeling, and sentiment classification using SVM. To address data imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied. Model performance was evaluated using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The results show that the SVM model achieved an accuracy of 77.00% without tuning and 75.15% after hyperparameter optimization, with improved balance across sentiment classes. These findings indicate that SVM is effective for sentiment classification of YouTube comments.

Fenomena Hollow Shell Effect Pada Aplikasi Finansial: Evaluasi Paradoksal User Experience Kaspro di Kalangan Pengemudi Maxim

Ilham Al Hafidz Muhyiddien — 2026-04-30

Digital transformation in the online transportation sector positions digital wallets not merely as optional payment tools, but as absolute work infrastructures for driver partners. This study aims to critically evaluate user satisfaction and experience with the KasPro application within the Maxim Driver Jabodetabek community. Given the high mobility and heterogeneity of the gig worker population, this research is positioned as an exploratory quantitative study. Data collection involved 99 respondents, determined using Slovin's formula with a 10% margin of error through purposive sampling techniques. Evaluation using the User Experience Questionnaire (UEQ) instrument revealed a sharp usability paradox. The Perspicuity scale achieved a highly satisfactory positive score (2.32), indicating an easily learnable interface. However, the other five crucial dimensions fell into the Bad category with extreme negative values: Attractiveness (-2.39), Efficiency (-2.42), Dependability (-2.50), Stimulation (-2.46), and Novelty (-2.35). Theoretically, this anomaly, where Perspicuity is inversely proportional to functional satisfaction, occurs due to the Hollow Shell Effect under conditions of forced adoption. For gig workers who have an absolute dependence on daily income liquidity, visual navigational ease instantly loses its significance when the system fails to meet basic utilities such as access speed and financial transaction security. This gap triggers passive resistance behavior, where users minimize operational interaction. The main contribution of this research is expanding human-computer interaction literature by validating the Hollow Shell Effect anomaly within the gig worker ecosystem, and proposing a design paradigm shift towards Reliability-First UX principles. Therefore, developers are recommended to prioritize backend infrastructure optimization through microservices architecture and e-KYC system automation, rather than simply carrying out visual aesthetic updates.

Peningkatan Kualitas K-Means Clustering Data Audio Musik Menggunakan Transformasi TableDC

Muhammad Aksa Hermawan — 2026-04-30

Clustering audio music data with high features typically suffers from performance degradation due to the curse of high dimensionality. A dataset with 518 classical K-Means features typically struggles to model nonlinear relationships between data. The purpose of this study is to analyze the implementation of the TableDC latent space transformation technique in the preprocessing stage before K-Means on the FMA Small dataset. This case study contains 8,000 songs with 518 audio features and is divided into eight music genres. The performance of K-Means on the original data is compared with that of K-Means on the latent space extracted by TableDC. The analysis is performed using several metrics such as the Silhouette Score, Davies-Bouldin Index, Calinski-Harabasz Index, Adjusted Rand Index, inertia or WCSS, and the number of iterations. The experimental results indicate a percentage improvement offered by the method. The Silhouette Score increased by 53 percent from the initial value of 0.0249 to 0.0382. Similarly, the ARI value increased from the initial value of 0.0876 to 0.0893. However, these absolute values remain very low, indicating that the formed cluster structures are still weak and substantially overlapping. In this case, the latent representation contributed to increasing the convergence efficiency from 63 to 47 iterations. The WCSS value also decreased from 3,433,413 to 20,628. However, unlike the two previous indicators, the linear-based DBI and CHI actually obtained better results compared to the initial model, which demonstrates the model's weakness in the context of conventional evaluation. Overall, the TableDC transformation has been shown to improve computational efficiency, but its performance has not fully resolved the issue of overlapping class separation.

Sistem Pemantau Siklus Haid Sebagai Media Manajemen Kesehatan Reproduksi Menggunakan Metode Forward Chaining dan Certainty Factor

Dimas Alva Rizki — 2026-04-30

The lack of understanding regarding the normal limits of physiological menstrual parameters leads to delayed detection of reproductive health disorders. Current conventional tracking applications generally focus on date prediction without analyzing accompanying symptoms. This research provides a technical contribution in the form of an expert system design for the early diagnosis of menstrual disorders based on four basic physiological variables: menstrual duration, cycle length, blood volume, and pain symptoms. The system is built using the Forward Chaining method to map the diagnostic inference flow, and the Certainty Factor (CF) to calculate the percentage of the expert's confidence level in the initial medical conclusion. Rule base validation was conducted with a general medical expert as a reference for early-stage screening (Amenorrhea, Oligomenorrhea, Polymenorrhea, Hypermenorrhea, Hypomenorrhea, Dysmenorrhea, and Normal). Black Box functionality testing shows that the system logic runs validly according to the static rule boundaries. Evaluation using the System Usability Scale (SUS) on 30 respondents resulted in a score of 83, indicating that the application has an excellent level of usability. As an early detection prototype, this system focuses on presenting diagnostic probabilities based on expert certainty, although continuous clinical validity testing using a Confusion Matrix remains necessary to measure medical accuracy comprehensively.

Identifikasi Tingkat Intensitas Opini dalam Analisis Sentimen Berbasis Aspek Menggunakan Enhanced Triplet Extraction

Gabriel Jimmy Richardo Chastelo B — 2026-04-30

Conventional sentiment analysis often overlooks variations in the intensity of opinions within text reviews. This is due to the limitations of the Aspect-Based Sentiment Analysis (ABSA) approach, which is restricted to three main triplet components. This study aims to develop and expand the Aspect-Sentiment-Opinion Triplet Extraction (ASOTE) framework to extract entity relationships and sentiment polarity by integrating opinion intensity detection. This study implements the ABSA approach by expanding the triplet structure into four components: aspect, opinion, intensifier, and sentiment (Enhanced Triplet). Data was collected via web scraping of Twitter (X) comments related to the Free Nutritious Meals program, which served as a case study to test the model’s ability to analyze public sentiment. The data then undergoes pre-processing and BIO Tagging, and is classified using a fine-grained sentiment approach to capture the nuances of emotional intensity in greater detail. A Transformer-based model, namely IndoBERT, was used to understand the context and intensity of meaning in the Indonesian language. Evaluation results on the test data show that the model achieved an accuracy of 88% and an average F1-score of 0.88 in sentiment polarity classification between entities, indicating strong model performance. These results demonstrate that providing a framework that is more sensitive to the intensity of opinions when classifying the nuances of public sentiment is a highly effective solution.

Implementasi Role-Based Access Control (RBAC) pada Sistem Monitoring Kenaikan Jabatan Fungsional Guru

Muhammad Rivaldi — 2026-04-30

As educators, functional position promotion (JaFung) serves as a means of advancing teachers’ careers as well as an indicator of their success in meeting established quality standards. However, in practice, this process still faces several challenges, such as teachers having difficulty monitoring the progress of their submitted proposals and institutions needing to contact teachers individually when issues arise. These problems indicate the absence of integration between progress monitoring and communication within a unified system. This study aims to propose a system model that integrates monitoring, communication, and RBAC-based security mechanisms. The Role-Based Access Control (RBAC) mechanism is used as a framework for managing interactions among actors across institutions and is designed using a scenario-driven role engineering approach to produce an RBAC model that adheres to the principle of least privilege. The system development method employed in this study is the Software Development Life Cycle (SDLC) using the Waterfall model. The results of black-box testing show that the developed system operates in accordance with its specifications and fulfills functional requirements. Access violation testing involving 10 unauthorized access attempts resulted in a violation rate of 0%, indicating that all unauthorized access was successfully prevented by the system. In addition, there was a 37.21% reduction in the number of permissions in the RBAC-implemented system, demonstrating the application of the least privilege principle. The proposed system not only facilitates document management and proposal progress monitoring but also enhances security and ensures more structured access management in the functional position promotion process.

Implementasi Scrum untuk Meningkatkan Adaptivitas Pengembangan Sistem Business Intelligence Pada Perusahaan Distributor Alat Kesehatan

Nursanti Novi Arisa — 2026-04-30

In the Industry 4.0 era, optimizing data utilization has become an important factor in enhancing organizational monitoring effectiveness and decision-making processes. PT Promedika Mitra Utama and PT Promedika Mitra Farma have digitalized various operational aspects, including employee activities, correspondence management, risk management, and weekly reporting. However, the generated data have not been optimally integrated to support managerial analysis. This study aims to design and implement a dashboard-based Business Intelligence (BI) system to improve monitoring effectiveness and managerial information accessibility. The development process includes performance metric identification, data collection and cleansing, data integration, and centralized data storage, with visualization implemented using Google Looker Studio. The Scrum method was applied to accommodate evolving variables and visualization requirements throughout iterative development and stakeholder feedback. The system development was completed in three sprints with a 100% backlog completion rate. Evaluation through sprint reviews and stakeholder validation demonstrated that the dashboard successfully accommodated changing requirements and supported a more systematic and integrated monitoring process. The resulting dashboard consists of three main reports, namely an integrated operational report and weekly reports for each company. The findings indicate the effectiveness of the Scrum approach in developing an adaptive BI system.

Sistem Pemesanan Restoran Berbasis QR Code dan Kitchen Display System

Muhammad Irzan — 2026-04-30

This research was motivated by the ongoing manual food ordering process at the Bakmi Ayam Bangka Chandra Restaurant, resulting in service delays, potential recording errors, and a lack of integration between customers, cashiers, and the kitchen. This situation impacts operational efficiency and customer satisfaction levels. The objective of this research was to design and develop a web-based food ordering information system integrated with the Kitchen Display System (KDS) to improve the speed, accuracy, and effectiveness of the restaurant's service process. The research method used was a qualitative approach, with data collection techniques including observation, interviews, and literature review. The system was developed using the Waterfall method, encompassing the stages of needs analysis, system design, implementation, and testing. The designed system allows customers to place orders independently by scanning a QR code, while order data is sent in real time to the kitchen and cashier without manual recording. The result of this research is an integrated ordering system that accelerates service flow, reduces the risk of communication errors, and provides more structured transaction recording. Implementing this system improves restaurant operational efficiency, enhances service responsiveness, and optimizes order management.

Identifikasi Penggunaan Chat GPT Pada Esai TOEFL Menggunakan Metode Long Short Term Memory

Karina Natasya Darmawan — 2026-04-30

The use of Artificial Intelligent (AI) technology is increasing along with technological developments. One of the technologies that is often used is Chat GPT (Generative Pre-trained Transformers). Chat GPT is an application used for many things such as source of information, write an essay, and answer TOEFL essay questions. Because of its easiness, people will excessively use this that can cause people to lose creativity because they do not understand the material context and rely too much on the AI text result, which poses academic risks. Teachers also have difficulty to distinguish between AI and human text writing. Therefore, this research is to identify whether TOEFL essay are result of human text or GPT. This research used the Long Short Term Memory (LSTM) method to identify the use of GPT in TOEFL essay. This research also used 3 different split data configurations to find the best results. This research consists of 2 TOEFL essay datasets with the same prompt and has total of 220 data samples. The LSTM method is a modification of algorithm Recurrent Neural Network (RNN) and part of Deep Learning. The LSTM method involves memory cell controlled by three gates, such as input gate, forgot fate, output gate, and the hidden state. The gates are used to decide and control the information added, deleted, and removed from memory cell. The results of this research is a system that can help teachers detect the use of GPT in TOEFL essay. This research successfully identified the use of GPT in TOEFL essay in a 70:30 data split configuration with a loss score of 25,07%, accuracy score of 89,83%, and prediction score of 64,32%. Therefore, it is hoped that this system can help teachers identify the use of GPT and facilitate the assessment of TOEFL essay.

Implementasi Business Intelligence Untuk Analisis Data Tingkat Kerawanan Kebakaran Berbasis Wilayah

Javier Alvino Alfian — 2026-04-30

This study aims to implement a Business Intelligence (BI) approach to analyze fire risk levels based on regional characteristics using open government data from Satu Data Jakarta. The dataset consists of 5,471 records for the period 2024–2025, including hazard, vulnerability, and capacity indicators at the neighborhood (RW) level. The methodology involves Extract, Transform, Load (ETL) using Python, data warehouse design with a star schema in PostgreSQL, OLAP-based analysis using SQL queries, and visualization through a web-based dashboard. The results indicate a significant increase in the proportion of high-risk areas from 16.62% in 2024 to 33.42% in 2025. However, this increase does not fully reflect actual changes in field conditions and may also be influenced by data distribution and the underlying risk classification system. Furthermore, the analysis reveals that fire risk is not evenly distributed but concentrated in specific regions, particularly in several districts of East Jakarta and South Jakarta, highlighting the importance of spatial-based approaches in determining mitigation priorities. This study utilizes pre-defined risk categories provided by the data source without performing predictive modeling or reclassification. The BI implementation not only integrates disparate data but also uncovers distribution patterns, risk trends, and regional priorities more systematically compared to conventional descriptive analysis. The findings contribute to supporting data-driven decision-making, especially in identifying priority areas for fire risk mitigation.

Prediksi Minat Belajar Siswa Berdasarkan Nilai Dan Intensitas Bermain Game Menggunakan Algoritma K-Nearest Neighbor

Karimah Agustin — 2026-04-30

This study aims to predict students' learning interest based on Islamic creed scores and their frequency of playing the Free Fire game using the KNN algorithm. The dataset used includes Islamic creed assignment scores, final exam (UAS) scores, game play time, frequency of play, and learning interest categories. The analysis process was carried out through several stages: data exploration, data cleaning and processing, selecting the best values, modeling, and disseminating the model results. The model was tested using the 5-fold cross validation method and confusion matrix to ensure that the prediction performance was adequate. The results showed that academic scores, especially Islamic creed final exam scores, had a greater influence on learning interest than the intensity of playing the Free Fire game. The KNN model with a k value of 6 produced an accuracy of 94.73% and a very small prediction error, as seen in the Confusion Matrix and Classification Report. These results indicate that the model is capable of working well in classifying students' learning interest into medium or high categories. This prediction model is expected to be used as a tool to understand student learning interest patterns in the school environment. The contribution of this research is the application of KNN-based machine learning methods in analyzing student learning interests, which were previously dominated by conventional approaches such as regression and questionnaires.

Dynamic Interdependence Between Altcoin Dominance and Ethereum Price: A Temporal Pattern-Based Analysis

Diva Ramadhani Ristiaji Putri — 2026-04-30

The cryptocurrency market exhibits dynamic and time-varying relationships driven by shifts in market structure and investor behavior. This study investigates the dynamic interdependence between altcoin dominance and Ethereum price, addressing the limitations of static correlation analysis by applying a temporal pattern-based approach. Using 1,416 daily observations from 2022 to 2025, the data are segmented into monthly periods to capture time-varying relationships. The analysis combines correlation, trend, and volatility metrics with pattern classification to identify recurring relationship structures across different market conditions. The results reveal a moderate negative correlation (r = ?0.48) at the aggregate level. However, the monthly analysis shows that this relationship is not stable over time, but instead varies across different market regimes. The relationship is dominated by inverse patterns (40.43%), followed by weak (38.30%) and positive (21.28%) patterns. From an economic perspective, the negative relationship can be explained by capital rotation dynamics within the cryptocurrency market. When altcoin dominance increases, market liquidity tends to shift from major assets such as Ethereum to a broader set of alternative tokens, leading to downward pressure on Ethereum prices. Conversely, during certain bullish periods, capital inflows can simultaneously strengthen both altcoin dominance and Ethereum price, resulting in positive relationships. These findings demonstrate that the relationship between altcoin dominance and Ethereum price is dynamic and context-dependent. The study highlights the importance of temporal segmentation and pattern-based analysis in capturing complex market behavior that cannot be explained by a single aggregate correlation measure.

Persistensi Artefak Telegram Web pada Memori Setelah Perubahan Sistem dengan Metode NIST SP 800-86

Sigit Puspito Wigati Jarot — 2026-04-30

The increasing incidence of cyberbullying on online communication platforms presents significant challenges for digital forensic investigations, particularly when perpetrators delete all message histories. Telegram Web, a browser-based messaging platform, produces volatile digital artifacts because its activity data is stored in system memory (RAM). This study aims to analyze the persistence of Telegram Web digital artifacts in volatile memory under six device condition variations using the NIST SP 800-86 framework, addressing a research gap in the quantitative evaluation of acquisition conditions for browser-based platforms. A cyberbullying simulation was conducted via Telegram private chat, generating 10 digital artifacts text messages, images, a document, and an audio file all subsequently deleted by the perpetrator. Memory acquisition was performed using Exterro FTK Imager under six conditions: immediately post-incident, sleep mode, hibernate mode, browser closed, browser closed with subsequent application use, and shutdown. Artifact identification employed keyword-based analysis on memory images. Results show that the first three conditions yielded 100% artifact recovery, as RAM preserved Chrome process data through DRAM self-refresh (ACPI S3) and byte-for-byte copying to hiberfil.sys (ACPI S4). Closing the browser reduced recovery to 40%, subsequent application use further reduced it to 10% due to zero-fill operations on reallocated memory pages, and shutdown produced 0% as all DRAM capacitor charges were lost. These findings demonstrate that artifact recovery rates are predictable from computer memory architecture, providing empirical guidance for digital forensic practitioners in web-based cybercrime cases.

Kombinasi Multi-Objective Optimization on the Basis of Ratio Analysis (MOORA) dan Simple Additive Weighting (SAW) Pada Sistem Pendukung Keputusan Seleksi Magang

Arya Fauzan Adima — 2026-04-30

The internship selection process at XZ University is still carried out conventionally, thus reducing time efficiency in decision making. The purpose of this study is to design a web-based Decision Support System (DSS) by applying a combination of Multi-Objective Optimization on the Basis of Ratio Analysis (MOORA) and Simple Additive Weighting (SAW) methods to increase the objectivity and efficiency of the selection process. The MOORA method functions to normalize and optimize alternative values as the basis for calculating SAW in internship selection, while the SAW method is used for normalizing and ranking alternatives based on weight criteria. This study uses a quantitative approach with a descriptive experimental method. Data were obtained through observation, interviews, and documentation. The criteria used include GPA, semester, collaboration time, video editing skills, camera operating skills, photography skills, videography skills, graphic design skills, creativity, attitude, discipline, responsibility. The results of data processing show that the five best alternatives with the highest preference values are A24 (100%), A3 (97%), A9 (96%), A5 (95%), dan A22 (93%.). These results show that the combination of the MOORA and SAW methods is able to provide the best alternative recommendations objectively and structured in the internship selection process.

Analisis Ternak Menggunakan K-Means Clustering Dalam Business Intelligence

Sulaiman Savero Sagi — 2026-04-30

This study aims to analyze livestock population patterns and classify regions in Central Java using data from 2020–2023 covering cattle, goats, and chickens from official sources. The Knowledge Discovery in Database (KDD) framework and K-Means Clustering were applied, with the optimal number of clusters determined using the Elbow Method and Silhouette Score. The results show that the optimal number of clusters is two (K=2), with a Silhouette Score of 0.328, indicating a relatively weak clustering structure with potential overlap. Despite this limitation, the results reveal meaningful segmentation when combined with Business Intelligence analysis. Cluster 0 represents regions with lower population but higher growth, while Cluster 1 represents regions with higher population but lower or negative growth. Further analysis indicates that the relationship between population, growth, and production is not linear, where high production does not necessarily correspond to strong growth. These findings highlight the importance of distinguishing between current production capacity and future growth potential, providing more informative insights for data-driven decision-making in livestock sector management.

Klasifikasi Hate Speech dan Offensive Language Menggunakan BERT dan Support Vector Machine

Muhammad Tirta Syakban — 2026-04-30

Hate speech and offensive language have become increasingly complex problems on social media, requiring classification approaches that can effectively capture linguistic context. While transformer-based models with end-to-end fine-tuning have become the dominant approach, the use of transformers as fixed feature extractors combined with classical machine learning algorithms remains relatively underexplored, particularly in benchmark settings such as HASOC 2021. This study aims to investigate the effectiveness of a feature-based transformer approach by combining embeddings from BERT and RoBERTa with Support Vector Machine (SVM) classifiers using multiple kernel configurations, including Linear, RBF, Polynomial, and LinearSVC. Experiments were conducted on Sub-task A and Sub-task B by comparing traditional feature-based methods (TF-IDF) with transformer-based embeddings. The experimental results show that RoBERTa embeddings consistently outperform other feature extraction methods. On the test dataset, the combination of RoBERTa and SVM achieves competitive performance compared to other systems in HASOC 2021. In Sub-task B, the optimal model achieves a Macro F1-score of 0.61, outperforming several BERT-based and classical baseline systems.These findings demonstrate that using transformer embeddings as fixed feature representations combined with optimized SVM classifiers can serve as an effective alternative to fine-tuning approaches, particularly in achieving more stable performance under class imbalance conditions. This study contributes by highlighting the potential of feature-based transformer methods as a flexible and competitive strategy for hate speech and offensive language detection.