Journal of Information Systems and Informatics

Bibliometric Analysis of Cybersecurity Research Trends in Bangladeshi Educational Institutions (2020-2025)

2025-09-22T09:41:06+07:00

This study provides a bibliometric analysis of cybersecurity research in Bangladeshi educational institutions from 2020 to mid-2025. Using data from the Scopus database and tools like R and VOSviewer, the results show a steady increase in research output, from 23 publications in 2020 to 77 in 2024, with projections for continued growth in 2025. Key research areas include network security, machine learning, deep learning, and blockchain technologies. Rajshahi University of Engineering and Technology has been a leading institution, with Md. Alamgir Hossain (State University of Bangladesh) being a prominent contributor, publishing 15 articles and accumulating 358 citations. International collaborations have enhanced Bangladesh's global standing in cybersecurity. These findings highlight Bangladesh’s increasing role in cybersecurity research, with implications for addressing local challenges and strengthening national cybersecurity resilience.

Enhancing the Security of Internet of Things Devices through Cybersecurity Framework

2025-09-22T08:16:46+07:00

This study focused on enhancing the protection of IoT devices by assessing the effectiveness of existing cybersecurity frameworks (CSFs), identifying gaps in advanced technology cyber-attack tactics, and developing a comprehensive cybersecurity framework for IoT ecosystems. Technological Acceptance and Zero Trust Security Theories guided the study. A cross-sectional research design and mixed-methods approach was adopted, while semi-structured interviews and Focus Group Discussions provided in-depth qualitative insights. For quantitative data, a questionnaire was used. A total of 93 respondents from HLIs, hospitals, and broadcasting media were selected using purposive and random sampling techniques. Descriptive and inferential statistics were employed to analyze quantitative data. For qualitative data, Atlas.ti 9.0 Desktop was used. The findings revealed cyber vulnerabilities are associated with the spread of imported unsecured IoT devices, user unawareness, and lack of effective cybersecurity frameworks tailored to emerging cyber threats from advanced technologies such as AI, 5G, Edge computing, and Autonomous Systems. In conclusion, a framework was designed to strengthen IoT device security by integrating best practices, policy implementation, and technological safeguards. The study recommends that imported IoT devices should be digitally coded to detect cyber risks and adopt multi-layered ECSF-IoT framework and strengthen end-user cybersecurity education in developing countries such as Tanzania.

LSTM Forecasting and K-Means Clustering for Passenger Mobility Management at Bus Terminals

2025-09-22T08:16:46+07:00

Rapid urban population growth has increased the need for efficient public transportation systems, particularly at bus terminals as major mobility hubs. To address operational challenges such as traffic congestion and limited infrastructure, this study proposes an innovative data-driven approach. A hybrid model is applied, integrating Long Short-Term Memory (LSTM) for passenger volume forecasting and K-Means Clustering for mobility pattern segmentation at the Jepara Bus Terminal. Monthly passenger data was utilized, and the K-Means method was applied to group monthly mobility patterns into three categories: low, medium, and high. The optimal cluster selection (k=3) was based on the highest Silhouette score of 0.785, providing clear seasonal insights. Analysis results indicate that September is the peak mobility period, while months like January and February fall into the low category. Furthermore, an LSTM model was trained to predict future passenger volumes. The model's performance was carefully validated and proven accurate, with a Mean Squared Error (MSE) of 0.0304 and a Root Mean Squared Error (RMSE) of 0.1745. These findings confirm that the model is reliable in capturing complex passenger movement patterns. Overall, this study concludes that the combination of LSTM and K-Means is an effective solution for supporting proactive decision-making. The results of this study can assist terminal managers in optimizing resource allocation and formulating more adaptive operational strategies, thereby contributing to the development of a more responsive and efficient intelligent transportation system.

Application of Life Simulation Games in Teaching Network Security and Cryptography

2025-09-22T08:16:46+07:00

Information security-related mathematical methods are used in the science of cryptography. A collection of methods that offer information security, cryptography is more than just a means of concealing messages. Using only presentation slides or video links at each meeting, the interaction between lecturers and students via SIPEJAR e-learning hinders the Network Security and Cryptography learning process at the State University of Malang (UM) Information Engineering (IT) Undergraduate Study Program. To help students learn more about the area of encoding using SIPEJAR, a game that explicitly explains cryptography was created using these several challenges as the background. The creation of a cryptographic life simulation game is intended to serve as a teaching and learning aid for lecturers and students. Students are expected to better understand related material in a learning atmosphere that is new, more interesting, opens the horizons of the mind, and is more investigative. After going through the equivalence partitioning testing process, in general this system produces a total percentage of 100% in system item test success in the testing process of the 6 item tests carried out and a respondent satisfaction percentage of 84.3%. Thus, the system is running according to the prototype design.

Factors Driving Internet Banking Adoption in Guyana: A Study of Developing Countries

2025-09-22T08:16:46+07:00

Internet banking across banking institutions has grown tremendously in popularity over the past two decades. Internet banking among customers remains a crucial challenge within the banking industry, especially in developing countries. As such, this research investigates the factors affecting internet banking adoption in Guyana by extending the Technology Acceptance Model (TAM) to include information quality, service quality, system quality and computer self-efficacy as predictor variables. The study evaluated hypotheses that these variables influence users’ perceived ease of use and perceived usefulness, which in turn affect actual usage of internet banking services. Data from 160 internet banking customers was collected and analysed using the Structural Equation Modelling (SEM) approach to test eight (8) hypotheses among constructs of the extended TAM model. The findings of the study suggest that service quality positively affects consumers’ perceived ease of use of Internet banking, while computer self-efficacy positively affects consumers’ perceived usefulness to adopt Internet banking. The findings also demonstrated that both perceived ease of use and perceived usefulness significantly impacted the actual usage of Internet banking. The findings of this study offer Guyanese banking institutions useful information, emphasizing the necessity of enhancing service quality standards and funding digital literacy programs to increase the adoption of online banking services.

Hybrid Cloud Architecture for Efficient and Cost-Effective Large Language Model Deployment

2025-09-22T09:40:18+07:00

Large Language Models (LLMs) have achieved remarkable success across natural language tasks, but their enormous computational requirements pose challenges for practical deployment. This paper proposes a hybrid cloud–edge architecture to deploy LLMs in a cost-effective and efficient manner. The proposed system employs a lightweight on-premise LLM to handle the bulk of user requests, and dynamically offloads complex queries to a powerful cloud-hosted LLM only when necessary. We implement a confidence-based routing mechanism to decide when to invoke the cloud model. Experiments on a question-answering use case demonstrate that our hybrid approach can match the accuracy of a state-of-the-art LLM while reducing cloud API usage by over 60%, resulting in significant cost savings and a ~40% reduction in average latency. We also discuss how the hybrid strategy enhances data privacy by keeping sensitive queries on-premise. These results highlight a promising direction for organizations to leverage advanced LLM capabilities without prohibitive expense or risk, by intelligently combining local and cloud resources.

A Comparative Analysis of Machine Learning Techniques and Explainable AI on Voice Biomarkers for Effective Parkinson’s Disease Prediction

2025-09-22T10:38:37+07:00

Parkinson's disease (PD) is a neurological movement disorder that remains difficult to diagnose, although it affects millions globally. Early diagnosis can lead to more effective and improved patient outcomes. Diagnosis through traditional methods is subjective and often lacks transparency, raising concerns about reliability. In this study, the CRISP-DM framework was applied to compare eight ML algorithms, including Random Forest and Support Vector Machine (SVM). Recursive Feature Elimination (RFE) was used to preprocess, balance, refine the data and find the eight most predictive vocal features. With 195 recordings coming from the UCI Parkinson’s Speech Dataset, which contains voice measurements from 31 individuals (23 with PD and 8 healthy controls), Random Forest (Entropy) had the best performance (????₁ = 96.6%, ROC AUC = 0.98). Explainable AI tools (SHAP and LIME) were integrated, allowing both global and instance-level understanding of model predictions thereby identifying measures of pitch variability (MDVP: RAP, spread1, PPE) as key predictors of PD. This research contributes to the practical deployment of reliable, transparent PD prediction tools in real-world medical settings, supporting early diagnosis and improved patient care. This raises the issue of the urgent need to detect PD early among Africa's aging populations to help protect the cultural heritage contained in the voices of the elders. this research contributes to the practical deployment of reliable, transparent PD prediction tools in real-world medical settings, supporting early diagnosis and improved patient care.Future work should embark on validating these findings over much more varied cohorts, integrating additional data modalities (e.g., gait, imaging), and enhancing model robustness. Real-time speech analysis-based tools, in the end, will allow remote screening, early intervention, and tailored care.

A Blockchain-Based Digital Library System Integrated with CryptoJS for Enhanced Security and Transparency

2025-09-22T11:46:15+07:00

In the context of digital library systems, blockchain presents a promising framework for enhancing the security, integrity, and transparency of operations such as book transactions, cataloging, and user authentication. Library systems face several challenges, including lack of transparency and security vulnerabilities. Previous research efforts have explored various centralized digital library management systems, but they often suffer from single points of failure and insufficient security measures. The methodology involves integrating blockchain technology using CryptoJS for advanced encryption and hashing, the backend was designed using PHP (Laravel), while the technologies used in the front end includes HTML, CSS and Javascript. The blockchain technology was implemented using Cryptojs which provides security by implementing AES encryption to safeguard user credentials and book transaction records, preventing unauthorized usage. The system was tested in a digital library environment and diverse user set, where results demonstrated enhanced data security and improved operational efficiency. The system is scalable and adaptable to academic, research, and public libraries, providing real-time verification of transactions and enhanced protection against unauthorized access. By combining blockchain’s immutability with strong encryption and modern web technologies, the platform delivers a secure, transparent, and future-ready solution for digital library management with 88% effectiveness. Findings indicate that the proposed blockchain-integrated system not only resolves existing issues in digital library management, but also introduces new opportunities for innovation, including real-time transaction verification and improved trust among users.

Ensemble Learning for Software Defect Prediction: Performance, Practicality and Future Directions

2025-09-25T10:42:36+07:00

Ensemble learning is a leading approach in software defect prediction (SDP), offering improved predictive performance on imbalanced and high-dimensional datasets. Despite growing research interest, persistent gaps remain in model interpretability, generalizability, and reproducibility, limiting its practical adoption. This paper presents a comprehensive analysis of 56 peer-reviewed studies published between 2020 and 2025, spanning both journal and conference venues. Findings show that ensemble methods, especially when combined with sampling, feature selection, or optimisation, consistently outperform single classifiers on important metrics such as F1-score, area under the curve, and Matthew correlation coefficient. Nonetheless, few studies incorporate explainability frameworks, effort-aware evaluation, or cross-project validation. Additionally, most models are static, rely on within-project testing, and depend on legacy datasets such as PROMISE and NASA, which limit external validity. Building on this synthesis, the review highlights future research priorities, including interpretable ensemble architectures, adaptive modelling, dynamic imbalance handling, semantic feature integration, and real-time prediction. Standardised benchmarks, transparent, scalable designs are recommended to bridge the gap between experimental performance and deployment-ready SDP solutions.

Enhancing News Similarity with Chunking Strategy and Hyperparameter Setting on Hybrid SBERT - Node2Vec Model

2025-09-25T11:02:26+07:00

The proliferation of online news necessitates accurate article similarity systems to combat information overload, yet models based solely on semantic content often ignore crucial structural context like news source and publication date. This research proposes and evaluates a hybrid embedding model that integrates semantic representations from Sentence-BERT (SBERT) with structural representations from Node2Vec. A series of quantitative experiments were conducted on the challenging, multilingual SPICED dataset to determine the optimal model configuration. Using Mean Squared Error (MSE) for evaluation, the results show that a per-paragraph chunking strategy yielded the best performance. This strategy's effectiveness was validated by the identical performance of an optimal fixed-size chunk (450 characters with a 64 overlap), a value that aligns closely with the dataset's average paragraph length. Furthermore, a community-focused (BFS-like) Node2Vec configuration (p=1.0, q=2.0, l=60) was identified as optimal for the structural component. Significantly, the final hybrid model (MSE = 0.1434) proved superior to both the purely semantic (MSE = 0.1449) and purely structural models (MSE = 0.2512). This study concludes that the fusion of content and context provides the most comprehensive and accurate representation for news similarity detection.

An Integrated Random Forest for Analyzing Public Sentiment on the “Makan Bergizi Gratis” Program

2025-09-25T11:16:40+07:00

The “Makan Bergizi Gratis” (MBG) Program is a public policy aimed at improving the nutritional quality of the community, particularly vulnerable groups. However, the success of this program is heavily influenced by public sentiment and perception. This research analyzes public sentiment toward the MBG program thru the social media platform X using an ensemble-based machine learning approach. The proposed framework integrates the Random Forest algorithm and compares it with four other ensemble models: AdaBoost, XGBoost, Bagging, and Stacking. A total of 3,417 tweets were analyzed using the TF-IDF method, both with and without stemming. The Random Forest model showed the best performance with an accuracy of 91.15% and an ROC-AUC of 95.46% on the data without stemming, consistently outperforming the other models. Additionally, a visual analysis of word frequency provides a strong indication of public opinion. These findings demonstrate the effectiveness of Random Forest in managing unstructured sentiment data and provide valuable insights for policymakers to monitor public responses and improve program implementation with greater precision.

Enhancing Coffee Leaf Rust Detection with DenseNet201Plus and Transfer Learning

2025-09-25T11:43:04+07:00

Coffee leaf rust (CLR) is a disease of coffee leaves caused by the fungus Hemileia Vastatrix, posing a major threat to global coffee production. Early and accurate detection is crucial for sustainable farming practices and disease management. This study proposes a novel deep learning approach that integrates DenseNet201Plus, an enhanced version of DenseNet201, with transfer learning to improve the accuracy and efficiency of CLR detection. DenseNet201Plus incorporates fine-tuned layers and optimized hyperparameters designed for plant disease classification, while transfer learning utilizes pre-trained weights from large-scale image datasets, enabling the model to adapt the characteristics of CLR images with limited training data. The model was evaluated on two datasets: the newly collected, high-quality Mbozi CLR dataset and the publicly available ImageNet CLR dataset, using accuracy, precision, recall, and F1-score. Results demonstrate that DenseNet201Plus achieved an accuracy of 99.0% on the Mbozi dataset, surpassing 97.78% obtained by the ImageNet Public dataset, with corresponding gains across all performance metrics. Results confirm that integration of DenseNet201Plus with transfer learning on the high-quality dataset significantly enhances CLR detection. The method outperformed several other baseline methods. The proposed approach offers a scalable, real-time detection solution for field deployment, supporting precision agriculture, enabling timely and targeted interventions.

Multi-Criteria Evaluation Based on MOORA for Improving Water Treatment Operations

2025-09-25T12:12:19+07:00

Access to clean and sustainable drinking water continues to be a significant concern, especially in areas with considerable variability in source quality. This study used the Multi-Objective Optimization on the Basis of Ratio Analysis (MOORA) approach to evaluate and rank 22 drinking water sources in Central Java, Indonesia, according to several physicochemical characteristics. The study process starts with the entry of sub-district, village, time, and laboratory result data, subsequently leading to the establishment of assessment criteria and their corresponding weights. Subsequent to the MOORA computations, rankings are produced and compiled into a detailed report. The results indicate that sources X21, X19, and X18 got the best ratings, signifying excellent water quality conditions, whereas X12 rated lowest, underscoring the necessity for focused action. In contrast to conventional evaluation methods, MOORA provides computational efficiency, clear prioritizing, and less subjectivity, facilitating consistent and reproducible multi-criteria evaluations. The results offer practical suggestions for enhancing water treatment processes, prioritizing resource distribution, and directing future incorporation of Internet of Things (IoT) monitoring for real-time assessment and adaptive management. This method integrates technical evaluation with pragmatic policy formulation, enhancing operational efficiency and promoting long-term sustainability in water delivery systems.

Impact of UI/UX on Shopee User Acceptance: A TAM Approach

2025-09-25T13:24:46+07:00

In the digital era, e-commerce platforms such as Shopee must continually improve their user interface (UI) and user experience (UX) to enhance user acceptance and competitiveness. This study analyzes the impact of UI/UX on user acceptance of the Shopee application using the Technology Acceptance Model (TAM), incorporating four variables: Perceived Usefulness (PU), Perceived Ease of Use (PEOU), Behavioral Intention to Use (BIU), and Actual System Use (ASU). A quantitative approach was applied, collecting data via questionnaire from a purposive sample of 90 active Shopee users in RT 002/07, Pela Mampang. Data were analyzed using SPSS 26, including validity, reliability, and hypothesis testing. The results show that PEOU significantly influences PU, while both PU and PEOU have a strong and significant effect on BIU, with PU demonstrating a slightly stronger influence. BIU also significantly affects ASU. These findings indicate that ease of use and perceived benefits are key drivers of user intention and actual usage behavior. The results provide practical implications for Shopee's design and development teams to prioritize enhancing ease of navigation, feature intuitiveness, and visual clarity to increase user engagement and system usage.

Securing EEG-based Brain-Computer Interface Systems from Data Poisoning Attacks

2025-09-25T13:46:38+07:00

Electroencephalogram (EEG)-based brain computer interface (BCI) is a widely used access technology to aid human-computer interactions. It enables communication between the human brain and external devices directly without the need for actuators such as human hands and legs. The BCI system acquires brain signals from an EEG device and uses machine learning (ML) algorithms to analyze and interpret the signals into actionable commands. However, EEG-based BCI systems are vulnerable to data poisoning attacks, which compromises the accuracy and security of the BCI system, and user safety. The objective of this paper is to protect the BCI systems against backdoor data poisoning attacks for reliable system operations. In this paper, a backdoor detect-and-clean mechanism, code named Bkd-DETCLEAN, to secure EEG-based BCI systems against data poisoning (backdoor) attacks is proposed using the Random Forest Classifier. Two models were designed, trained and validated on both clean and poisoned dataset respectively. The results of experiments on two benchmark EEG datasets shows that our solution achieves a detection accuracy of 98.5%, effectively identifying poisoned samples with a little below 5% false positive rate. Continued data cleaning iterations restored the poisoned training set, resulting in an overall system accuracy improvement from 78.9% to 93%. Based on these results, the proposed model sustained high detection and cleaning efficiency with different poisoning rates, underscoring the effectiveness of the machine learning driven proposed model in ensuring that brain signal integrity is not compromised. The proposed mechanism is also applicable in other areas including healthcare and medical data protection, protecting fraud detection models in financial systems, ensuring the integrity of sensor data in industrial control systems, protecting against user data manipulation in recommender systems, etc.

Performance Comparison of Sentiment Classification Algorithms on SIGNAL Reviews Using SMOTE

2025-09-25T14:16:14+07:00

Public service apps like SIGNAL are widely used to provide public access to information and vehicle tax payments. However, diverse user reviews highlight the need to evaluate public perception through sentiment analysis. Selecting an appropriate classification algorithm is crucial to ensure accurate results, particularly when dealing with imbalanced review data. Therefore, This study examines the comparative performance of four algorithms Naïve Bayes, Random Forest, Decision Tree, and SVM in analyzing the sentiment of 36,000 user feedback obtained from Google Play Store. The dataset underwent preprocessing, feature extraction using TF-IDF, and class balancing using SMOTE. Model evaluation was conducted using accuracy, precision, recall, and F1-score. The findings indicated that Random Forest performed the best overall performance (accuracy 91.04%, F1-score 94.80%), followed by Naïve Bayes (accuracy 89.89%, F1-score 93.38%), SVM (accuracy 89.22%, F1-score 93.02%), and Decision Tree (accuracy 88.40%, F1-score 92.31%). These findings indicate that Random Forest is highly effective for balanced datasets, while SVM and Naïve Bayes offer competitive precision for applications prioritizing accuracy in positive class detection. The output of this study can be applied practically by developers and related institutions in optimizing public service applications and by applying Random Forest algorithm to gain actionable insights for optimizing features and aligning services more closely with user needs.

Improving IT Governance Maturity at Universitas Sebelas Maret Using COBIT 2019

2025-09-25T14:29:36+07:00

This study evaluates and improves the IT governance maturity of the Directorate of ICT at Universitas Sebelas Maret using the COBIT 2019 framework. The evaluation was driven by increasing IT complexity, resource inefficiencies, and low risk management capability. A case study approach applied COBIT 2019 domains to assess practices and identify gaps, with data gathered through interviews, observations, and document analysis. Significant deficiencies were found in six key processes. The highest gap score is APO12 (Managed Risk) at 1.89, followed by DSS04 (Managed Continuity) at 1.88, DSS01 (Managed Operations) at 1.75, APO14 (Managed Data) at 1.74, DSS05 (Managed Security) at 1.57, and the lowest is APO01 (Managed I&T Framework) at 1.27, with all domains targeting a maturity level of 3. Results indicate current maturity scores fall below expectations, highlighting the need for systematic improvement. A phased strategic plan was developed for short, medium, and long-term priorities, aligned with resources and organizational needs. The study demonstrates that structured implementation of COBIT 2019 can enhance governance alignment, improve risk control, and ensure sustainable ICT performance, providing a roadmap for future IT governance at the university.

Sentiment Analysis of Public Service Using Naïve Bayes Classifier

2025-09-25T15:08:57+07:00

Public administrative service quality is a crucial factor in citizen satisfaction. This study analyzes sentiment in public service reviews using a text mining approach with the Naïve Bayes Classifier method. The dataset was collected from citizen feedback on online platforms regarding public administrative services. Preprocessing steps included tokenization, case folding, stopword removal, and stemming. The Naïve Bayes algorithm with Laplace smoothing was applied for classification, and performance was evaluated using accuracy, precision, recall, and F1-score. The experiment resulted in an accuracy of 91.2%, precision of 90.3%, recall of 89.7%, and F1-score of 90.0%. The analysis revealed that Service Speed obtained an average score of 3.21, indicating a moderate level of citizen satisfaction in that aspect. These findings suggest that while the Naïve Bayes method is effective for sentiment classification, its greatest value lies in providing actionable insights for public service improvement. Specifically, policymakers can prioritize addressing delays in service speed through simplified procedures, improved staffing, and digital innovation, while maintaining strengths such as officer politeness and effective complaint handling. By leveraging sentiment analysis, public institutions can continuously monitor citizen feedback, identify problem areas, and implement evidence-based strategies to enhance service quality and strengthen public trust.

NLP-Based Sentiment Analysis of Alfagift and Klik Indomaret Application Reviews: A Comparative Study

2025-09-26T09:37:03+07:00

Amid competition for online shopping applications, Alfagift and Klik Indomaret compete for the same market share. This study aims to analyze and compare user reviews of both applications using sentiment analysis based on Natural Language Processing (NLP) with the E-Servqual approach, focusing on Efficiency and System Availability indicators, to determine the advantages and disadvantages of each application and provide a basis for service improvement, strategic decision making, and reference for users in choosing online shopping applications that suit their needs. Methods include data collection, data grouping, data processing, selecting analyzed samples with consensus, and data analysis to describe user perceptions of the quality of service of each application. The results showed that on the positive side, both apps experienced an increase in efficiency although not significant, with gradual improvements in user experience. Alfagift showed improvements in technical responsiveness and ease of use, while Klik Indomaret was relatively stable with a simple user experience. On the negative side, efficiency issues still arise consistently and impact user perception. Alfagift often faces access and login issues, while Klik Indomaret tends to be slow when accessing various features. These findings reflect that despite year-on-year improvements, both apps still face technical challenges that need to be resolved to improve the overall quality of digital services.