JOIV : International Journal on Informatics Visualization
Not a member yet
786 research outputs found
Sort by
Enhanced Adverse Drug Event Extraction Using Prefix-Based Multi-Prompt Tuning in Transformer Models
Get PDFExtracting mentions of adverse drug events and relationships between them is crucial for effective pharmacovigilance and drug safety surveillance. Recently, transformer-based models have significantly improved this task through fine-tuning. However, traditional fine-tuning of transformer models, especially those with many parameters, is resource-intensive, memory-inefficient, and often leaves a gap between pre-training and downstream task-specific objectives. Soft prompting is a lightweight approach that updates a trainable prompt to guide task-specific fine-tuning, showing comparable performance to traditional fine-tuning for large language models on simple tasks. However, its effectiveness on complex tasks like token-based sequence labeling requiring multiple predictions for a single input sequence remains underexplored, particularly in multi-task settings. In addition, using holistic prompts in multi-task learning settings may be biased to other subtasks. Additionally, some prompt tokens hurt the model prediction. This study proposes a prefix-based multi-prompt soft tuning method with attention-driven prompt token selection for tuning transformer models on multi-task dual sequence labelling for concept and relation extraction. We experimented with BERT and SciBERT models using frozen and unfrozen parameter strategies. Our approach achieved state-of-the-art performance on the n2c2 2018 and TAC 2017 datasets for adverse drug event extraction, with multi-prompt tuning in unfrozen models surpassing traditional fine-tuning. Moreover, it outperforms the largest clinical natural language processing model, GatorTron, on the n2c2 2018 dataset. This research highlights the potential of soft prompts in efficiently adapting large language models to complex downstream NLP tasks
Convolutional Neural Networks-Based For Predicting Aerodynamic Coefficient Of Airfoils At Ultra-Low Reynolds Number
Get PDFMany applications, including airplane design, wind turbines, and heat transmission, use symmetric or asymmetric airfoils. Engineers employ these airfoil shapes to optimize performance and efficiency. Each airfoil has a unique set of aerodynamic coefficients that must be calculated to maximize the airfoil design. Engineers utilize numerous ways to calculate coefficients, such as lift and drag. One of the methods is the prediction method, which effectively reduces time and cost. This study's training dataset is obtained from particle-based numerical computation using the Lattice Boltzmann Method (LBM). Then, Convolutional Neural Networks (CNN) are used as a prediction method to get the aerodynamic coefficients of airfoils for lift and drag based on two different Reynolds numbers. In CNN, airfoil geometry representation is essential. The Signed Distance Function (SDF) was used to convert airfoil geometry into RGB pictures. On the other hand, the SDF method cannot explain different flow conditions; in this case, it is represented by the Reynolds number (Re). Therefore, we propose a Text-based Watermarking Method (TWM) to differentiate between Re = 500 and Re = 1000. Each airfoil representation was trained and tested to generate each prediction model using a modified LeNet-5. The computation results show that using CNN with TWM on SDF to define the Reynolds numbers could predict the lift and drag coefficients with varying angles of attack. Future research can focus on generalizations to different aerodynamic aspects and practical applications in complex scenarios
Evaluating Mixed Reality Technology for Enhancing Art Pedagogy
Get PDFThe lack of interest among students in studying art, particularly the traditional Indonesian art form of batik, poses a significant challenge for educational institutions. Despite its cultural significance, the education sector lacks effective strategies to introduce and enhance students' interest in batik within the art curriculum. Several consequences can arise if the education sector fails to implement strategic measures to address this issue promptly. This could lead to a gradual erosion of cultural heritage and a loss of artistic traditions passed down through generations, and students may miss out on valuable opportunities for self-expression and cultural exploration. This study addresses this issue by leveraging mixed reality and gamification in a batik creation application. This innovative approach not only enhances the pedagogy of art education but also aims to revive cultural interest. The study employs Software Testing and PIECES to evaluate user experiences, emphasizing user comfort and smooth interactions. By assessing the application with tools like Unity Profiler and Hololens 2 performance testing, the study ensures an optimal user experience, contributing to the broader goal of preserving Indonesia's cultural heritage through innovative and accessible educational solutions. The results fall within the range of 4.04 to 4.24, categorizing user satisfaction as "satisfied" and the application running at an optimal 60 frames per second (FPS). This implies that users responded positively to the application, indicating that implementing mixed reality technology in batik learning provides a satisfying experience
Predicting Battery Storage of Residential PV Using Long Short-Term Memory
Get PDFSolar power panels, or photovoltaic (PV), have recently grown rapidly as a renewable alternative energy source, especially since the increase in the basic electricity tariff. PV technology can be employed instead of the state electricity company to reduce the electricity used. Indonesia is one of the countries that have great potential in producing electricity from PV technology, considering that most of Indonesia's territory gets sunlight for most of the year and has a large land area. Considering the benefits of PV technology, it is necessary to carry out predictive monitoring and analysis of the energy generated by PV technology to maximize energy utilization in the future. The Internet of Things (IoT) and cloud computing system was developed in this research to monitor and collect data in real-time within 27 days and obtained 7831 data for each parameter that affects PV production. These data include data on the light intensity, temperature, and humidity at the location where the PV system is installed. The feature selection results using Pearson correlation revealed that the light intensity parameter significantly impacted the PV production system. This research used the Long Short-Term Memory (LSTM) method to predict future PV production. By tuning hyperparameters using 3000 epochs, the resulting RMSE value was 171.5720. The results indicated a significant change in the RMSE value compared to 100 epochs of 422.5780. This model can be applied as a forecasting system model at electric vehicle charging stations, given the increasing use of electric vehicles in the future.    Keywords— Forecasting; energy; Photovoltaic; LSTM; Internet of Thing.Â
Analysis of Pneumonia on Chest X-Ray Images Using Convolutional Neural Network Model iResNet-RS
Get PDFPneumonia, a prevalent inflammatory condition affecting lung tissue, poses a significant health threat across all age groups and remains a leading cause of infectious mortality among children worldwide. Early diagnosis is critical in preventing severe complications and potential fatality. Chest X-rays are a valuable diagnostic tool for pneumonia; however, their interpretation can be challenging due to unclear images, overlapping diagnoses, and various abnormalities. Consequently, expedient, and accurate analysis of medical images using computer-aided methods has become crucial. This research proposes a Convolutional Neural Network (CNN) model, specifically the ResNet-RS Model, to automate pneumonia identification. The Contrast Limited Adaptive Histogram Equalization (CLAHE) technique enhances image contrast and highlights abnormalities in pneumonia images. Additionally, data augmentation techniques are applied to expand the image dataset while preserving the intrinsic characteristics of the original images. The proposed methodology is evaluated through three testing scenarios, employing chest X-ray images and pneumonia dataset. The third testing scenario, which incorporates the ResNet-RS model, CLAHE preprocessing, and data augmentation, achieves superior performance among these scenarios. The results show an accuracy of 92% and a training loss of 0.0526. Moreover, this approach effectively mitigates overfitting, a common challenge in deep learning models. By leveraging the power of the ResNet-RS model, along with CLAHE preprocessing and data augmentation techniques, this research demonstrates a promising methodology for accurately detecting pneumonia in chest X-ray images. Such advancements contribute to the early diagnosis and timely treatment of pneumonia, ultimately improving patient outcomes and reducing mortality rates
E-Nose for Piston Ring and Cylinder Block Condition Detection of Motorcycle Engine Based on MyRIO LabVIEW Programming
Get PDFThis study has created a system capable of identifying the condition of the piston ring and cylinder block of a 4-stroke motorcycle engine using petrol or similar through exhaust emissions. Multisensory gas, sensitive to changes in CO, CO2, NOx, and HC gas elements and compounds, is installed as an input to the exhaust channel and integrated using LabVIEW programming on the NI myRIO module. Multisensory data is processed using the FFT and the backpropagation method to classify whether the piston rings and engine cylinder block are in good or damaged condition. Tests have been carried out on motorbikes with piston rings and engine cylinder blocks that are in good, damaged, or unknown condition. During the test, the target error value for motorcycles with piston rings and engine cylinder blocks in good or damaged condition is less than 1%. The system can distinguish the condition of the piston ring and cylinder block of a motorcycle engine that is 100% optimal and 100% damaged with an error of 0% compared to the compression test method, and the maximum error is 20% Compared to the technician's manual method. Ten motorcycles were randomly tested in unknown conditions; 50% were in good condition, and 50% were damaged. For further development, an electronic nose system can detect engine combustion conditions and damage to cylinder rings and 4-stroke motorbike engine blocks based on exhaust emissions
The Impact of Online Learning on NDUM Students During COVID-19
Get PDFOne of the impacts of the COVID-19 outbreak was the closure of numerous education facilities, including schools and universities. Due to the closing of these institutions, the method used for teaching and learning changed from physical face-to-face lecturing to online contactless learning. This helps curb the spread of infections while ensuring that teaching and learning continue as usually as possible. However, questions arise not only about the effectiveness of online learning but also about the impact of online learning on education stakeholders, namely students and educators. This study aims to assess the effects of the lockdown during COVID-19 on National Defense University of Malaysia (NDUM) students. A link pointing to a custom-built questionnaire was forwarded to students through email and WhatsApp. At the end of the survey period, 445 students responded to the questionnaire. The simple percentage distribution was employed to evaluate the student's learning status and their expectations. Based on the analysis, during the lockdown, students faced issues involving technical, time management, social interactions, and surrounding (home-related) issues. In contrast, during the lockdown, students were also keen to learn new technological skills and favorable towards the ability to replay lectures and class materials. These valuable insights on the impact of online learning on students are essential due to the advancement of technology in education, not only in Malaysia but in other nations as well
Personalized Learning Models Using Decision Tree and Random Forest Algorithms in Telecommunication Company
Get PDFIn response to the rising popularity of online training, this study addresses the crucial need for effective assessment methods at PT XYZ. The research focuses on developing a comprehensive solution through a data visualization dashboard and a machine learning model. The data visualization dashboard, created using Tableau, provides an interactive platform for exploring training data. It offers valuable insights into employees learning progress and needs, empowering them to monitor their advancement and identify areas for improvement effectively. Simultaneously, a machine learning model was developed using Python and Google Collab, employing decision trees and random forest algorithms. The model exhibited promising results with an accuracy rate of 69% for decision trees and 70% for random forests, indicating its proficiency in predicting skill groups. Furthermore, the study rigorously evaluated the dashboard and machine learning model using a 20% holdout dataset, affirming their effectiveness. The dashboard, deployed on a web server, ensures accessibility to all PT XYZ employees, enhancing user experience and engagement. Notably, the dashboard's user-friendly interface allows employees to actively participate in their learning journey, while the machine learning model generates personalized training recommendations based on their progress and needs. In summary, this research provides a practical and innovative solution to the challenge of online training assessment at PT XYZ. By combining data visualization techniques and machine learning algorithms, the developed tools significantly enhance the efficiency and effectiveness of training programs. These findings contribute valuable insights into online training assessment methodologies and pave the way for improved learning experiences in the digital age
Development of an IoT-Based Egg Incubator with PID Control System and Web Application
No full textThe rapid development of technology significantly impacts various aspects of life, including the field of livestock farming. The advancement of technology is expected to enhance the rate and effectiveness of production, particularly in the hatching of chicken eggs or chick breeding. The existing technology relies on manual on/off systems and manual monitoring, hindering successful egg-hatching rates and percentages. Therefore, this research aims to explain the development of an automated egg incubator using a Proportional Integral Derivative (PID) control system with hypertuning parameters, as well as temperature and humidity monitoring, along with a protection system based on voltage sensors, all integrated with the Internet of Things (IoT). The PID control is employed to regulate the temperature of the egg incubator, ensuring stability according to the predetermined set point temperature. The IoT system in this study comprises an ESP32 node as a microcontroller connected to a sensor, using Firebase and User app for monitoring the egg incubator. The study employed PID control with parameter values Kp=10, Ki=3, and Kd=8. The research yielded time-efficient egg incubation and prevention of turning delays. The DHT21 sensor achieved a 90% success rate in detecting room temperature (38°C) and humidity (77%-84%) within the incubator, while PID control effectively maintained the target temperature. The ACS712 sensor accurately detected current in the heater, power supply, and motor. The Kodular application can display sensor readings. The future implication is developing a more adaptive PID method toward changes and nonlinear dynamics.
Boosting Vehicle Classification with Augmentation Techniques across Multiple YOLO Versions
Get PDFIn recent years, computer vision has experienced a surge in applications across various domains, including product and quality inspection, automatic surveillance, and robotics. This study proposes techniques to enhance vehicle object detection and classification using augmentation methods based on the YOLO (You Only Look Once) network. The primary objective of the trained model is to generate a local vehicle detection system for Malaysia which have the capacity to detect vehicles manufactured in Malaysia, adapt to the specific environmental factors in Malaysia, and accommodate varying lighting conditions prevalent in Malaysia. The dataset used for this paper to develop and evaluate the proposed system was provided by a highway company, which captured a comprehensive top-down view of the highway using a surveillance camera. Rigorous manual annotation was employed to ensure accurate annotations within the dataset. Various image augmentation techniques were also applied to enhance the dataset's diversity and improve the system's robustness. Experiments were conducted using different versions of the YOLO network, such as YOLOv5, YOLOv6, YOLOv7, and YOLOv8, each with varying hyperparameter settings. These experiments aimed to identify the optimal configuration for the given dataset. The experimental results demonstrated the superiority of YOLOv8 over other YOLO versions, achieving an impressive mean average precision of 97.9% for vehicle detection. Moreover, data augmentation effectively solves the issues of overfitting and data imbalance while providing diverse perspectives in the dataset. Future research can focus on optimizing computational efficiency for real-time applications and large-scale deployments