JOURNAL OF INFORMATION SCIENCE AND TECHNOLOGY

Development of a deep learning-based early Alzheimer’s disease screening system using MRI for the Thai healthcare context

2026-05-19T21:04:31+07:00

Thailand’s transition into an aging society has increased the prevalence of Alzheimer’s disease, making early-stage screening essential for improving patient outcomes. This paper proposes a deep learning-based screening system for early detection of Alzheimer’s disease using magnetic resonance imaging (MRI), tailored to the context of the Thai public healthcare system. A total of 6,400 MRI images has been used and classified into four categories: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. Data preprocessing and augmentation techniques have been applied to enhance model performance. Four models have been evaluated, including the proposed Custom CNN, MobileNetV2, ResNet50, and VGG16. The results show that the proposed Custom CNN achieves the highest accuracy of 97.12%, with strong generalization and stable training performance. The model has also effectively reduced misclassification in early-stage categories. The proposed system demonstrates strong potential as a practical screening tool for early Alzheimer’s detection in Thailand, particularly in primary healthcare settings.

Application of Data Mining Techniques for Developing an Information System for Water Demand Forecasting

2026-06-24T12:47:02+07:00

This study aims to compare the performance of water demand forecasting models in Thailand using data mining techniques under the CRISP-DM framework. The models include Linear Regression, Artificial Neural Network, Random Forest, and Gradient Boosted Trees. Monthly water consumption data from 2014 to 2023 were obtained from the national open data platform and divided into training, validation, and testing sets at a ratio of 80:10:10. RapidMiner was used for model development. Performance evaluation was conducted using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The results show that Linear Regression achieved the best performance with an MAE of 188,158.96 m³ and an RMSE of 310,809.74 m³, which were lower than those of the other models and demonstrated high stability. A web-based prototype system was also developed to visualize forecasting results through dashboards, supporting water production planning and distribution management in alignment with Sustainable Development Goal 6.

Hybrid Emotion and Satisfaction-Aware Fuzzy Inference System (ES-FIS) for Predicting Tourist Revisit Intention

2026-06-10T13:53:50+07:00

Tourist revisit intention is influenced by both cognitive evaluations and emotional experiences; however, existing models typically treat these factors independently and rely on precise numerical inputs that may not reflect the subjective nature of travel perception. This study develops a Hybrid Emotion–Satisfaction Fuzzy Inference System (ES-FIS) designed to model such ambiguity while maintaining interpretability through linguistic rules. The model integrates six structured satisfaction dimensions from 415 Thai visitors to Japan with affective polarity extracted from 20,491 TripAdvisor reviews via a probabilistic mapping approach. All variables were transformed into fuzzy linguistic terms and processed using a Mamdani-type inference engine to generate both a continuous revisit score and categorical outputs. Empirical evaluation indicates that the ES-FIS achieves strong predictive performance (Accuracy: 0.78, F1-score: 0.74) with results aligned with ground-truth Likert ratings (r = 0.71; RMSE = 0.48). Furthermore, PCA-based clustering reveals three distinct cognitive–affective traveler segments. These findings suggest that combining satisfaction and emotional indicators provides a more comprehensive representation of revisit intention than relying on either component alone. The study contributes to an explainable AI (XAI) framework that captures the inherent imprecision of tourist perception and offers practical value for service personalization and CRM-oriented segmentation in tourism analytics.

Low-Resource Language Text Generation Performance Comparison: LLaMA 3.1-13B-Instruct vs GPT-4-mini with Dataset Augmentation

2026-06-16T11:17:50+07:00

Large Language Models (LLMs) have demonstrated remarkable capabilities in text generation, yet their effectiveness for Indonesian, a Low-Resource Language (LRL), remains underexplored. This study systematically compares LLaMA 3.1-13B-Instruct and GPT-4-mini for Indonesian text generation using an augmented dataset designed to mitigate data scarcity. Three augmentation strategies—back-translation, synonym substitution via mBERT embeddings, and paraphrasing through GPT-4-mini—were employed to expand lexical and syntactic diversity. Quantitative results show that GPT-4-mini achieves higher BLEU (0.55 vs 0.52), ROUGE-L (0.64 vs 0.61), METEOR (0.57 vs 0.54), lower Perplexity (11.5 vs 12.8), and higher mBERTScore (0.90 vs 0.88) compared to LLaMA, indicating stronger lexical and semantic alignment. Conversely, LLaMA exhibits greater lexical diversity (Distinct-2 = 0.34 vs 0.31). Human evaluation involving four native raters confirms that GPT-4-mini excels in fluency (4.5 vs 4.1), coherence (4.4 vs 4.0), and relevance (4.4 vs 4.3), while LLaMA slightly surpasses in factual accuracy (4.5 vs 4.2). These findings highlight the complementary strengths of the models—GPT-4-mini for fluent and coherent generation, and LLaMA for factual precision and lexical richness—demonstrating the positive impact of data augmentation in improving Indonesian LLM performance.

Development and Comparison of YOLOV8 Models for Plastic Classification via Digital Image Processing

2026-06-16T22:37:29+07:00

Currently plastic problem has caused a lot of problems and has a lot of effected to Thailand and tends to have more plastic waste every year. Researcher points that having effective plastic waste management is essential by reusing plastic which can help the environment. Now Thailand still uses the legacy management to manage the plastic waste by using human which leads to inefficient result plastic is include with local waste which were burned, buried or leak to the sea. This research proposes the solution of plastic management to recycle plastic by using image using Deep learning with the data of PET, HDPE, PVC, LDPE, PP, PS and Other then training the model using YOLO. I use YOLOv8 so AI can separate type of plastic precisely and speedy. From the experiment’s result each YOLO version has potential to separate the type of plastic this research tested the performance of YOLOv8 with different version founded that YOLOv8n has given the best result. This research shows that the potential of AI can help with plastic waste management and can lead to innovation for environment.

Comparison of Machine Learning Model Performance in Classifying the Geographical Origin of Emeralds

2026-04-10T15:46:54+07:00

This research aims to 1) construct machine learning models for classifying the geographical origins of emeralds from four sources—Colombia, Zambia (Kafubu), Zambia (Musakashi), and Afghanistan; and 2) compare the performance and accuracy of these models in origin classification. The study followed the six-phase CRISP-DM data mining methodology. The dataset was divided into training and testing sets in an 80:20 ratio, and a SMOTE technique was applied to address data imbalance issues. Feature selection was performed using the One-Way ANOVA technique combined with the Pearson correlation coefficient, resulting in 11 optimal elemental features. Four supervised machine learning algorithms were employed to construct the classification models: Random Forests, Support Vector Machines, Naïve Bayes, and k-Nearest Neighbors. Model performance was evaluated using a confusion matrix. The results revealed that the Random Forests model achieved the highest classification performance with 100% accuracy, followed by the Naïve Bayes and k-Nearest Neighbors models, both attaining an accuracy of 99.08%. The Support Vector Machines model produced the lowest accuracy at 97.25%. These findings demonstrate that machine learning techniques are highly effective tools for supporting the classification of emerald origins.

Development of an Interactive Holographic Narrative Experience with Choice-Based Decision using Fictional Characters and Air Gesture Technology: A Case Study of the Yellow Building, Vajira Hospital Museum

2026-06-16T10:58:44+07:00

This research aims to design and develop an advanced interactive holographic display system to enhance the medical history learning-oriented museum experience at the Yellow Building, Vajira Hospital Museum. The project introduces three innovative features to tackle challenges associated with touch-based interfaces in hygienic environments and the limitations of conventional exhibitions. First, it employs a 45-degree plane reflection hardware architecture utilizing top-down projection to generate high-fidelity volumetric visuals at eye level. Second, it incorporates a choice-based branching narrative delivered by fictional digital characters acting as virtual guides, promoting visitor engagement. Lastly, it implements 100% touchless interaction through Ultraleap sensors and Touch Free software, positioned at a 45-degree angle via Snap Mounts for precise "Air Cursor" navigation. We want to make a working digital exhibition prototype that clearly shows the Results indicate that most participants historical importance of the Yellow Building while also making it easier for people to remember information and enjoy their visit. This study also lays out an important plan for how to use Natural User Interfaces (NUI) in modern museums, carefully balancing immersive user experiences with strict public health safety standards.

Explainable Analysis of Security Information and Event Management (SIEM) Event Logs for Event Profiling and Preliminary Triage

2026-06-19T00:07:54+07:00

ABSTRACT - Security Information and Event Management (SIEM) systems generate large volumes of heterogeneous event logs, imposing substantial investigation workloads on Security Operations Centers(SOCs). This study proposes an explainable framework for SIEM event analysis and preliminary triage that emphasizes interpretability and operational applicability without requiring labeled training data. The framework integrates descriptive and temporal analysis, structural association analysis using Spearman correlation, K-means clustering with TF-IDF features to construct interpretable event profiles, and explainable triage scoring based on standard SIEM indicators (severity, priority, confidence, aggregated counts). Experimental evaluation using operational SIEM logs (n=3,490 events) demonstrates 93.8% workload reduction while retaining 98.97% of high-severity events. Validation with labeled attack data (489 SQL injection attempts, 301 bruteforce attacks) achieves 98.2% detection rate (776/790 attacks correctly prioritized) with only 1.8% false negatives, substantially outperforming severity-only (78.5%) and severity+priority (89.1%) baselines. The results confirm that unsupervised event profiling combined with explainable triage scoring effectively distinguishes genuine security incidents from routine monitoring events. The framework provides a practical, transparent analytical baseline for SOCs environments that supports analyst decision-making under realistic operational constraints.