JOIV : International Journal on Informatics Visualization
Not a member yet
786 research outputs found
Sort by
Maximizing Prospective Students through Instagram-Driven Influence Maximization for University Branding
Full text linkIn many developing countries, private universities face considerable financial challenges due to their limited sources of income. Unlike public universities, which receive substantial government subsidies, private institutions must rely primarily on tuition fees and student enrollment to sustain their operations. This financial dependency pushes private universities, particularly in Indonesia, to invest heavily in branding and marketing to attract prospective students. Social media platforms, especially Instagram, have emerged as crucial tools in these efforts. Instagram enables universities to engage with potential students directly, providing a space for showcasing academic programs, campus life, and student achievements. This paper’s goal is to explore the potential of Instagram-driven influence maximization (IM) techniques for enhancing university branding. Traditionally, these techniques are employed in product marketing to increase consumer engagement and sales, but their application in higher education remains relatively unexplored. We aim to address this gap by examining three promising IM methods that private universities can adopt: first, leveraging Instagram’s broadcast channel feature to communicate effectively with large audiences in real time; second, selecting key opinion leaders (KOLs) from influencer marketplace platforms to enhance credibility and broaden outreach; and third, optimizing the use of targeted hashtags to increase discoverability and engagement. As a result, we provide a review of recent evidence supporting the effectiveness of these strategies in higher education marketing. By adopting these techniques, private universities can improve their digital presence, enhance brand awareness, and increase student enrollment. IM with uncertainty is a challenging educational landscape for future research
Multi-Head Voting based on Kernel Filtering for Fine-grained Visual Classification
Full text linkResearch on Fine-Grained Visual Classification (FGVC) faces a significant challenge in distinguishing objects with subtle differences within intra-class variations and inter-class similarities, which are critical for accurate classification. To address this complexity, many advanced methods have been proposed using feature coding, part-based components for modification, and attention-based efforts to facilitate different classification phases. Vision Transformers (ViT) has recently emerged as a promising competitor compared to other complex methods in FGVC applications for image recognition, which are mainly capable of capturing more fine-grained details and subtle inter-class differences with higher accuracy. While these advances have shown improvements in various tasks, existing methods still suffer from inconsistent learning performance across heads and layers in the multi-head self-attention (MHSA) mechanisms that result in suboptimal classification task performance. To enhance the performance of ViT, we propose an innovative approach that modifies the convolutional kernel. Our method considerably improves the method's capacity to identify and highlight specific crucial characteristics required for classification by using an array of kernels. Experimental results show kernel sharpening outperforms other state-of-the-art approaches in improving accuracy across numerous datasets, including Oxford-IIIT Pet, CUB-200-2011, and Stanford Dogs. Our findings show that the suggested approach improves the method's overall performance in classification tasks by achieving more concentration and precision in recognizing discriminative areas inside pictures. Using kernel adjustments to improve Vision Transformers' ability to differentiate somewhat complicated visual features, our strategy offers a strong response to the problem of fine-grained categorization
Incremental Learning Approaches for Dermoscopic Image Classification in Teledermatology
Full text linkThis study investigates the application of incremental learning techniques to enhance the classification of skin diseases in dermoscopic images. The research aims to develop a model capable of continuous adaptation to new data while retaining previously acquired knowledge. Two datasets were utilized: acne images and the HAM10000 dataset comprising various skin lesions. The methodology involved initially training a ResNet-18 convolutional neural network on 1,052 samples across eight classes, followed by an incremental learning phase incorporating 800 additional data points. Rigorous preprocessing steps were implemented to ensure data quality, including cropping, resizing, and normalization. Results demonstrate that the base model achieved 87% accuracy on the test set, which improved to 90% after the incremental learning process. Detailed analysis revealed significant improvements in precision, recall, and F1-scores for several skin disease classes, notably for challenging categories such as Basal Cell Carcinoma (bcc) and Dermatofibroma (df). Confusion matrix analysis and Grad-CAM visualizations provided insights into the model's decision-making process and its focus on clinically relevant features. The study also implemented a Streamlit application to demonstrate real-time classification capabilities and the system's adaptability in a simulated clinical environment. These findings have potential clinical applications, particularly in teledermatology systems where adaptive algorithms can accommodate new dermatological data over time. The study highlights the potential of incremental learning in creating accurate, adaptable, and clinically relevant AI models for skin disease classification in evolving medical practices
Addressing Challenges and Enhancing Sustainability in the Food Supply Chain Management for the Malaysian Armed Forces Based on IoT Technologies
Full text linkThe critical nature of the food supply chain issues within the Malaysian Armed Forces necessitates careful consideration to establish a well-structured and organized sustainable food supply. The primary source of frustration arises from the contractor's failure to adhere to contractual obligations, resulting in inadequate supplies, delivery delays, and provisions that do not meet the specified requirements. These shortcomings indirectly impede the management process. The aim of this paper is to identify the relationship between delivery handling, quality control, condition of storage, food supply chain management, and contract management towards the quality of military fresh rations. It is also focusing on improving food supply chain management in MAF, especially the quality of military fresh rations. In addition, this study proposes potential solutions to address these issues, providing a clear path for improvement. The research methodology for this study will employ a qualitative approach. The primary data will be gathered via questionnaire surveys and subsequently analyzed using SPSS. Finally, the study concludes with some recommendations for future research, highlighting areas for further investigation
2.5D Face Recognition System using EfficientNet with Various Optimizers
Full text linkFace recognition has emerged as the most common biometric technique for checking a person's authenticity in various applications. The depth characteristic that exists in 2.5D data, also known as depth image, is utilized by the 2.5D facial recognition algorithm to supply additional details, strengthening the system's precision and durability. A deep learning approach-based 2.5D facial recognition system is proposed in this research. The accuracy of 2.5D face recognition could be enhanced by integrating depth data with deep learning approaches. Besides, optimizers in the deep learning approach act as a function for adjusting the properties, like learning rates and weights in the neural network, which can minimize the overall loss of the system and further enhance performance. In this paper, several experiments have been conducted in two versions of EfficientNet architectures, such as EfficientNetB1 and EfficientNetB4, using different optimizers, including Adam, Nadam, Adamax, RMSProp, etc. Various optimizers are compared to find the most suitable optimizer for the system. The Face Recognition Grand Challenge version 2 (FRGC v2.0) database was utilized in this research. This research aims to increase the 2.5D face recognition system’s effectiveness and efficiency by implementing deep learning approaches. Based on the experimental result, a deep learning algorithm enhances the system's accuracy rate. It also proves that the EffifientNetB4, using Adam optimizer, gained the highest accuracy rate at 97.93%
Maize Leaf Disease Identification with Large and Lightweight Convolutional Neural Models
Full text linkTo minimize yield losses in maize plantations, control measures that include early leaf disease detection are essential. In this study, we evaluated extensive and lightweight convolutional neural network (CNN) models to accurately classify maize diseases from leaf images. To achieve a high image classification performance, existing deep learning approaches often use large models that require substantial computational resources. Simpler and lightweight models provide faster inferences but at the expense of lower accuracy in prediction performance. To improve maize leaf disease classification performance on the lightweight SqueezeNet model, the response-based knowledge distillation method was evaluated for model training. In response-based knowledge distillation, the logit output from the last layer of the large model is used in the loss function to train the lightweight model. This enables the lightweight model to learn from the knowledge of large and complex models, thereby improving its predictive accuracy while maintaining a simpler architecture and faster inference. A six-class maize disease dataset was prepared using two publicly available datasets. The dataset was used to train and evaluate the selected large and lightweight models. The large and lightweight model demonstrated high classification accuracy when trained till 40 epochs. The trained SqueezeNet model showed promising performance for accurately identifying various maize leaf diseases with an accuracy of 96.68%. When the model is trained with the response-based knowledge distillation method, the test accuracy improves to 97.13%. Such lightweight models with high accuracy can facilitate the deployment on resource-constrained devices
Enhancing Motoric Impulsivity Detection in Children through Deep Learning and Body Keypoint Recognition
Full text linkQuantifying motoric impulsivity in pediatric settings is crucial for safeguarding children and for devising effective intervention strategies. Existing quantitative techniques, such as accelerometry, have been utilized to assess it, but they often prove insufficient for accurately differentiating impulsive movements from regular ones. Conventional assessment methods are frequently used and rely on subjective assessments, which hinders the accurate characterization of impulsive behavior. To address this research gap, our study introduced an innovative objective approach using computer vision and deep learning techniques. We utilized MediaPipe to track precise body movement data from a child. The data were then analyzed using a Bidirectional Long Short-Term Memory (Bi-LSTM) network to process sequential information and recognize patterns indicative of impulsivity. Our approach successfully distinguished impulsive movements, marked by rapid changes in position and inconsistent movement velocities, from typical behavioral patterns with an accuracy rate of 98.21%. This research demonstrates the effectiveness of combining computer vision and deep learning to measure motoric impulsivity more precisely and impartially than prevailing qualitative techniques. Our model quantifies behaviors, enabling the development of improved safety protocols and targeted interventions in educational and recreational settings. This research has broader implications, suggesting a framework for future studies on pediatric motion analysis and behavioral assessment
A Hybrid Machine Learning Approach Utilizing PCA and ICA Features for Stress Classification
Full text linkElectroencephalographic (EEG) signal-based personal identification systems have notable advantages and disadvantages. These systems heavily rely on the stability of EEG signals, which several factors. One of the primary factors affecting the stability of EEG signals is an individual's emotional state. Among emotional states, stress significantly impairs people's ability to perform daily tasks. This research aims to identify stress levels, classified as low (2) and high (1), using features from Independent Component Analysis (ICA) and Principal Component Analysis (PCA). Machine learning methods, including Decision Tree, k-Nearest Neighbors (k-NN), Naive Bayes, Support Vector Machine (SVM), and Ensemble techniques, are employed to classify the stress levels. The dataset comprises 40 EEG recordings from the Stroop color-word test, and the data is split using a random holdout function with a ratio of 80% for training and 20% for testing. This study examines the most effective features for identifying stress levels and compares the performance of various machine learning models. The experimental results demonstrate that PCA is the most effective feature extraction method, achieving an average accuracy of 0.718 in stress level classification. Among the machine learning models tested, the Ensemble method performs the best, achieving an accuracy of 0.770 when using PCA features and 0.745 with ICA features. This study highlights the importance of selecting optimal features and machine learning techniques for improving stress detection in EEG-based systems. Further improvements in classification accuracy may be achieved by incorporating additional physiological signals or refining feature extraction techniques
A Study on the Development of Data Literacy Content Framework for Elementary School Students
Full text linkThis study aimed to develop a data literacy content framework for elementary school students to establish a foundation for systematic data literacy education. The research was conducted through a literature review and analysis of the 2022 Revised National Curriculum of Korea. Based on Ridsdale et al. (2015)'s framework, components of data literacy suitable for elementary students were derived, and curriculum achievement standards related to data literacy were analyzed to develop the content framework. The research identified eight key components of data literacy: understanding data, data collection, data evaluation, data organization, data analysis, data visualization, data-driven decision-making, and data ethics. Curriculum analysis revealed that science (36.3%) and social studies (32.7%) subjects contained the highest proportion of data literacy elements, with grades 5-6 (63.2%) including more achievement standards than grades 3-4 (36.8%). The developed framework is categorized into three domains: knowledge and understanding, processes and skills, and values and attitudes. It considers grade-level hierarchy by focusing on basic concepts and simple functions for grades 3-4, while emphasizing complex concepts and higher-order functions for grades 5-6. This study contributes to supporting systematic data literacy education in elementary schools by providing a content framework that considers students' cognitive developmental stages and is expected to foster future core competencies through practice-centered education. Further research is needed to verify practical applicability, develop teaching and learning methods, and strengthen connections between school levels
Prediction Analysis of Greeting Gestures Based on Recurrent Neural Networks
Full text linkHuman activity recognition, such as rehabilitation, sports, human behavior, etc., is developing rapidly. A Recurrent Neural Network (RNN) is a practical approach to human activity recognition research and sequential data. However, studies on recognizing human activities rarely study culture, including greeting gestures. And studies seldom use small datasets when employing the RNN approach, as they typically utilize large amounts of data to conduct such studies. This study aims to predict greeting gestures from Japan and Indonesia with limited data. This study proposes and compares six RNN architecture methods, including Long Short-Term Memory (LSTM), Bidirectional RNN (BRNN), Gated Recurrent Unit (GRU), Vanilla RNN (VRNN), Deep RNN (DRNN), and Hierarchical RNN (HRNN), which have been modified with regularization to handle overfitting. We evaluate using Mean Squared Error (MSE), Root Mean Squared Error (MSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). The experimental results show that LSTM has the best MSE, RMSE, and MAE values, with MSE of 0.0773479, RMSE of 0.2781149, and MAE of 0.2402451, while GRU has the best R² value of 0.0267571. The conclusion of this study indicates that LSTM and GRU are more suitable than other models for solving this problem. Therefore, it can be beneficial for future research to address the challenges of small data and overfitting in sequential data and human activity recognition, particularly in the context of greeting gestures. Future work can utilize data augmentation, proper parameter selection, and incorporate data from multiple individuals to enhance the accuracy of the model