International Journal of Advances in Data and Information System
Not a member yet
161 research outputs found
Sort by
A Usability Evaluation of an E-Commerce Checkout System Using SUS and Usability Testing
Full text linkThis study focuses on evaluating and improving the usability of the XYZ Checkout System, which previously exhibited several issues that negatively affected transaction completion, including unclear information hierarchy, inconsistent interface behaviour, and inefficient checkout flows. To address these challenges, the research adopted a User-Centered Design (UCD) approach supported by a mixed-method evaluation strategy, combining the System Usability Scale (SUS), usability testing, and open-ended questions. A between-subject research design was employed, involving two independent groups of participants. The initial as-is evaluation was conducted with 35 users and produced an average SUS score of 69.57, indicating marginal usability and the presence of notable interaction barriers. Based on the findings, a high-fidelity to-be prototype was designed using established usability heuristics and UCD principles. The redesigned interface was subsequently evaluated by a different group of 35 users, resulting in a significantly improved average SUS score of 96.64, classified as Excellent with high acceptability. Overall, the results demonstrate that the proposed design improvements successfully enhanced interface consistency, clarity of interaction, and checkout efficiency, emphasizing the critical role of systematic usability evaluation in optimizing e-commerce checkout systems
Digital Surveillance in Information System Research: A Hybrid Systematic Literature Review and Bibliometric Analysis
Full text linkThe advancement of digital technology has promoted the application of digital surveillance in many fields, including public health and social risk management. Nonetheless, little research has comprehensively mapped the field of digital surveillance studies. This paper seeks to investigate the publications, scientific structures, and significant topics of digital surveillance research through a systematic literature review integrated with bibliometric and thematic analyses. Records retrieved from Scopus, 2009-2025resulted in 286 articles that fulfilled the inclusion criteria. Bibliometric analysis was performed using VOSviewer to identify publication trends, main scientific fields, and relationships between keywords. Meanwhile, thematic analysis using NVivo was used to reveal research themes and sub-themes. The results showed that digital surveillance publications had increased significantly since 2020. Social sciences, medicine, and computer science dominated the publications. Thematic analysis revealed that the main research themes focused on the application of digital surveillance in public health, data and information system management, and the use of digital technology. However, issues of governance, ethics, and privacy protection were still underexplored. These findings contribute theoretically by mapping the landscape of digital surveillance research and, practically, by helping academics and practitioners identify opportunities for research and policy development that pay greater attention to ethical and digital governance aspects
A Heterogeaneous Dataset–Driven Ensemble Learning Framework for Malicious URL Detection
Full text linkModern cyberattacks are increasingly associated with phishing campaign, malware distribution, and website defacement, which are often delivered through malicious Uniform Resource Locator (URL) originating from diverse source. This paper examine malicious URL detection using an ensemble learning framework evaluated on large scale heterogeneous dataset composed of URL aggregated from multiple public threat intelligence source. The dataset include benign, phishing, malware, and defacement URL, thereby reflecting real world variability in attack pattern and data distribution. Three ensemble based classifier, namely Decision Tree (DT), Random Forest (RF), and Gradient Boosting (GB), are evaluated with respect to detection accuracy and computational efficiency. In addition to classification performance, this study present a detailed analysis of training and detection time in order to identify most suitable model for practical deployment. Experimental results indicate that the DT model achieves a training time of 4.14 seconds with macro and weighted accuracies of 94.11% and 91.71%, respectively, and a per category detection time of 0.2162 seconds. The RF model attains macro and weighted accuracies of 93.64% and 90.94%, with training and detection times of 9.73 seconds and 0.2420 seconds, respectively. Although the GB model exhibits the longest training time of 45.38 seconds, it achieves the fastest per category detection time of 0.2151 seconds. Despite its comparatively lower overall accuracy of 92.48% for macro averaging and 89.42s% for weighted averaging, the rapid inference capability of GB makes it a strong candidate for real time malicious URL detection in heterogeneous cybersecurity environments
Improving Precision in Small Area Proportion Estimation Using Logit Transformation: A Case of Internet Utilization in Papua’s Regencies, Indonesia (2021)
Full text linkThe development of technology and communication reflects economic growth, with internet usage serving as one key indicator. While this indicator is generally available at the national or provincial level, reliable estimates at the regency level remain limited. Small Area Estimation (SAE) methods can address this gap by integrating survey data with census or administrative records. However, the basic SAE model may be less suitable for proportions, particularly in rare cases, due to violations of the normality assumption. This study shows that applying a logit transformation within the SAE framework improves the precision of proportion estimates. Using internet usage in Papua, Indonesia, as a case study, the results demonstrate that the logit-transformed SAE model outperforms both direct survey estimates and the basic SAE model.
Enhancing AI-driven Cybersecurity Awareness Smart Consultant using RAG Method with Hybrid Knowledge Based
Full text linkThe rapid advancement of Artificial Intelligence (AI), particularly in the realm of Large Language Model (LLM), holds significant potential for addressing the escalating issue of cyberattacks that exploit users\u27 insufficient cybersecurity awareness. This study involves the design and development of a prototype cybersecurity awareness smart consultant, leveraging AI through the Retrieval Augmented Generation (RAG) method. This approach integrates hybrid knowledge derived from both user-specific internal cybersecurity documents and internet resources, thereby enhancing the validity of system responses and mitigating the risk of hallucinations. The prototype was evaluated using the Answer Accuracy Score (AAS) method, based on Black Box Testing and human evaluation, across four cybersecurity-related questions, yielding an average score of 0.925, accompanied by comprehensive analysis and discussion. The findings indicate that the system\u27s response accuracy improves when knowledge is synthesized from both internal document resources and internet sources. Future research may focus on incorporating deliberative thinking to further enhance system performance in generating responses
Hybrid Relevance and Sentiment Classification of Indonesian Gold Tweets Using Machine Learning for Market Risk Signal Extraction
Full text linkThis study proposes a hybrid relevance–sentiment classification framework to analyze public opinion on physical Antam gold from Indonesian Twitter data and to support exploratory market-risk signal extraction. Tweets were collected during February–November 2025, after preprocessing and text-normalized deduplication, 1,271 unique tweets were retained. The approach combines weak supervision (rule-/lexicon-based silver labels) with TF-IDF-based machine learning in two stages: (1) relevance classification to separate tweets genuinely discussing physical Antam gold from non-relevant contexts (e.g., ANTM stock/capital-market discussions), and (2) two-class sentiment classification (positive vs negative) applied to relevance-filtered tweets. Random Forest achieved the strongest relevance performance (Accuracy = 0.984; macro-F1 = 0.943; 5-fold CV macro-F1 = 0.928 ± 0.033). For sentiment classification, performance was moderate and close across models; the most stable model under cross-validation (Logistic Regression/Naive Bayes) was used for downstream aggregation. Sentiment outputs were aggregated into a monthly sentiment index for descriptive comparison with gold prices; the observed association was weak, indicating that the index is better interpreted as a risk-perception proxy rather than a direct price predictor
Empirical Performance of E2E Frameworks in React-Vue SPAs Using DIA
Full text linkModern web applications increasingly adopt Single-Page Application (SPA) architectures to enhance the user experience through client-side rendering and dynamic content loading. However, these characteristics introduce significant challenges for automated end-to-end (E2E) testing, including asynchronous DOM manipulation, complex state management, and timing synchronization issues. This study presents a comprehensive empirical comparison of three prominent E2E testing frameworks—Selenium WebDriver, Cypress, and Playwright—across React and Vue-based SPAs. Using a quantitative experimental approach, 25 standardized test cases were executed 15 times each across Chrome, Firefox, and Edge, for a total of 270 testing sessions. Performance evaluation focused on four key metrics: execution time, success rate, CPU usage, and memory consumption. Results demonstrate that Playwright achieved the fastest execution time (56.25 seconds on React-Chrome), while Selenium exhibited superior resource efficiency with the lowest memory consumption (196.59 MB on Vue-Chrome). The Distance to Ideal Alternative (DIA) multi-criteria decision analysis method identified Playwright-Chrome as optimal for React applications (DIA score: 0.886715) and Selenium-Chrome for Vue applications (DIA score: 0.908237), indicating that framework selection should be context-dependent based on application characteristics and deployment requirements. This research supports the conclusion that no universal "best" testing framework exists, underscoring the importance of evidence-based, application-specific tool selection in software quality assurance
Comparative Evaluation of Convolutional Neural Network Full Learning Model with Transfer Learning (VGG-16) for Coffee Bean Roasting Level Classification
Full text linkIndonesia is the 3rd largest coffee producing country in the world in 2022-2023 with coffee production reaching 11.85 million bags per 60 kg of coffee. One of the important processes in coffee production is roasting because the roasting level of coffee beans can affect the taste and aroma of coffee. The problem faced is that the process of assessing the level of coffee roasting is traditionally carried out through visual observation by an expert (roaster). This method produces a subjective level of assessment and requires high skills and experience, making the assessment of the level of coffee roasting less efficient and prone to human error. Therefore, in this study the author aims to develop a Convolutional Neural Network (CNN) model for the classification of the level of coffee bean roasting that can achieve better and faster accuracy. In this study, the author compared two CNN architecture approaches for the classification of the level of coffee bean roasting. The first approach is full learning with an architecture consisting of three convolution layers. The second approach is transfer learning based on the VGG-16 model. From the results of the analysis, it is known that the full learning model has a better level of accuracy and a faster running time than the VGG-16 transfer learning. The CNN full learning model for coffee bean roasting level classification is able to classify the coffee bean roasting level, with an accuracy of 98.75% and a running time of 856 ms per step. The application of CNN for coffee roasting level classification can provide benefits such as improving quality control and reducing the level of subjectivity of a roaster in assessing the roasting level of coffee beans
Ambidextrous Cloud Governance Approach to Enhance TelCo\u27s Digital Transformation Using COBIT 2019 Traditional and DevOps
Full text linkCloud computing plays a crucial role in accelerating digital transformation within the telecommunications sector by enhancing operational efficiency, scalability, and service innovation. However, TelCo faces difficulties in aligning its cloud adoption with effective governance, particularly in ensuring continuous service delivery and resilience. This study proposes a cloud governance framework based on the ambidextrous integration of COBIT 2019 Traditional components and DevOps Focus Area. Employing a Design Science Research methodology, data were collected through semi-structured interviews guided by a structured questionnaire and triangulated with internal documents until data saturation was achieved. Governance and Management Objectives were prioritized using the ambidextrous COBIT 2019 lens, supported by regulatory guidelines from the SOE Minister No. PER-2/MBU/03/2023 and the ICT Minister No. 5/2021, as well as relevant prior studies. The analysis highlighted DSS04: Managed Continuity as the most critical focus area. A capability gap assessment identified vacant leadership roles, overlapping responsibilities, and the lack of Infrastructure as Code (IaC) implementation in cloud services. Recommended improvements include formal assignment of leadership positions, clarification of IT responsibilities, and the adoption of IaC practices. These enhancements are expected to raise the capability maturity of DSS04 from 3.25 to 3.87, representing a 0.62 increase in business continuity readiness. This study contributes to research by extending ambidextrous governance theory to cloud continuity management and provides practical guidance for telecommunications providers seeking to improve operational resilience and manage risks effectively in their digital transformation efforts
A CNN-Based Information System for Balinese Dance Classification with Hyperparameter Optimization
Full text linkBalinese dance classification presents challenges due to limited datasets, complex postures, and the lack of real-world implementation. Existing studies often focus on model development while overlooking deployment aspects. This research proposes a lightweight Convolutional Neural Network (CNN) designed for Balinese dance classification and compares it with MobileNetV2, ResNet50, and VGG16 using consistent training settings. Data augmentation was applied to enhance generalization, and training epochs were optimized based on model convergence. The proposed CNN achieved a validation accuracy of 99.00%, with a precision of 92.55%, recall of 89.88%, and F1-score of 91.1%, using only 590 thousand trainable parameters and the fastest inference time of 476 milliseconds. Although others pretrained model, MobileNetV2 slightly outperformed in some metrics, the proposed model offered a better tradeoff between performance and efficiency. The trained model was deployed in a web-based application, demonstrating practical usability. This work supports the preservation of Balinese dance through accessible and efficient AI integration