Jurnal Nasional Pendidikan Teknik Informatika : JANAPATI

Virtual Smart School: A Blended Learning Approach for Schools in Papua’s 3T Regions

2025-02-08T07:27:35+00:00

Traditional classroom-based teaching methods are still used in schools, particularly in Papua’s 3T (frontier, outermost, undeveloped) areas. This highly relies on the teacher-centered method and limits the completion of the curriculum due to time constraints imposed by the schedule. This research aims to design a web-based virtual smart school as a blended learning solution for schools in Papua’s 3T areas. The virtual smart school learning media application is developed using the ADDIE model, a five-step process encompassing analysis, design, development, implementation, and evaluation. This research results in a virtual smart school application that can increase student motivation, enable independent learning, evaluate student learning progress, provide quick access to learning materials, and facilitate interactions between teachers and students at any time or location. The research concludes that virtual smart schools can become a blended learning solution to improve the quality and equity of education in underdeveloped areas of Papua.

A Security Enhancement to The Secure Mutual Authentication Protocol for Fog/Edge

2024-11-01T09:11:32+00:00

A secured mutual authentication protocol (SMAP Fog/Edge) has been developed for fog computing. The protocol provides secure mutual authentication which lightweight and efficient for fog computing environments. However, based on AVISPA’s verification this protocol has been found to be vulnerable to man-in-the-middle attacks due to the absence of an authentication scheme between authentication server and the edge user. The attacks are carried out assuming that the public key of the fog server is not distributed over a secure channel. We propose to modified this protocol to enhance the security of SMAP Fog/Edge and make it resistant to man-in-the-middle attacks. The proposed protocol is revalidated using the AVISPA tool to determine whether the vulnerability still exists. Additionally, we suggest a mechanism that utilizes encryption and digital signatures to substitute the secure channel for distributing the public key of the fog server and authenticating edge users by the authentication server.

Balinese Script Handwriting Recognition Using CNN and ELM Hybrid Algorithms

2025-01-14T03:02:06+00:00

One of the foundational scripts used in Balinese culture is the Balinese script, known as “Aksara Bali”. In its writing, Aksara Bali follows specific rules regarding distinctive stroke shapes that must be carefully maintained to preserve authenticity and readability. This study proposes the use of a hybrid algorithm combining Convolutional Neural Network (CNN) and Extreme Learning Machine (ELM) to recognize handwritten Balinese script characters. The preprocessing stage includes dataset splitting, rescaling, data augmentation, batch size adjustment, and visualization of class distribution. The training stage utilizes the Adam Optimizer to enhance model accuracy. Using 1,691 images of various Balinese script characters, the dataset is divided into an 80:10:10 ratio for training, validation, and testing. Experimental results show that the best accuracy achieved is 91%, indicating that the CNN-ELM hybrid model effectively recognizes Balinese script characters.

Analyzing User Experience and Satisfaction in the B-Block Game-Based Assessment

2025-03-17T02:23:29+00:00

Game-based assessment (GBA) has developed as an innovative education method, including learning basic arithmetic operations. This study aims to analyze user experience and satisfaction using B-Block, an assessment-based game for basic arithmetic operations. The study involved 94 junior high school students with an age distribution of 12-13 years old and varying levels of gaming experience. The research used descriptive statistical analysis, validity and reliability test, Pearson correlation test, and multiple linear regression to identify factors influencing user satisfaction and continuance usage intention. The analysis showed that B-Block has good usability and educational benefits, with user satisfaction being the most dominant aspect. Validity and reliability tests confirmed that most variables were valid and reliable (Cronbach's Alpha > 0.7), except Errors, which had lower reliability (α = 0.632). Pearson correlation shows that Perceived Usefulness has a strong relationship with satisfaction (r = 0.784), while user satisfaction contributes significantly to continuance intention (r = 0.694). Multiple linear regression revealed that perceived usability and perceived usefulness were the main factors influencing user satisfaction, while confirmation and satisfaction had the most effect on continuance intention. The findings confirm that the gameplay's usability and perceived usefulness are key in increasing user satisfaction while matching the experience with initial expectations, and user satisfaction contributes to continued use.

Prediction of Total Weight of Octopus Cyanea Using Multiple Linear Regression Method

2024-10-04T22:27:09+00:00

Fisheries Improvement Programs (FIPs) rely on data to offer recommendations for sustainable fishing practices. The octopus cyanea FIP in East Nusa Tenggara faces difficulties in data collection, particularly the total weight of the octopus, as the heads are often removed before landing. This is because the head's contents can cause rapid spoilage and blackening due to the ink. However, these contents are also used as bait. Understanding the total weight is crucial for linking it to gonad weight data to determine the octopus's maturity level. In this study, two models were developed to estimate the total weight of an octopus using known data through Multiple Linear Regression. The most accurate model used total length and body weight without the head contents as predictors, with a Mean Absolute Error (MAE) of 27.97 grams, indicating an average error of this amount in the predictions. The model's fit was assessed with an R2-Score of 0. 983, suggesting a strong correlation with the actual data. Additionally, T-test results indicate no significant statistical difference between the predicted and actual weights. This research aims to provide an alternative method for estimating the total weight of octopuses to support the Octopus FIP in Flores, East Nusa Tenggara.

Cost-Effective Parkinson’s Disease Diagnosis Through IoT-Based Finger Tapping and Real-Time Machine Learning Classification

2025-01-09T00:31:49+00:00

Parkinson's disease (PD) is a progressive neurological condition that significantly impacts motor functions, including finger tapping (FT). This study aims to develop a cost-effective, real-time, easily implementable, IoT-enabled electronic health record (EHR)-integrated FT analysis system capable of remotely detecting PD with high accuracy. The study investigates the use of peak amplitude, the Internet of Things (IoT), and various machine learning classifiers to detect PD through FT pattern analysis on a smartphone application. K-Nearest Neighbors, Convolutional Neural Networks, Support Vector Machines, and Logistic Regression exhibited 100% accuracy, while Naïve Bayes and Decision Trees (DT) had accuracies ranging from 71% to 92%. All classifiers had an Area Under the Curve (AUC) value of 1, except DT with an AUC value of 0.75. Data from healthy control (HC) and PD groups showed normal distribution, as determined by the Shapiro-Wilk test (p > 0.05), except for the HC group at peak 10. Significant differences were found between the PD and HC groups, as evidenced by the T-test and U-test (p < 0.05), with Pearson's r <= -0.79 and Spearman's rho = -0.58, indicating strong agreement between machine learning classification and neurologist evaluation. The proposed IoT-based approach demonstrated high diagnostic accuracy, cost-effectiveness, real-time monitoring capabilities, easy implementation, scalability for telemedicine, and accessibility to EHR during the COVID-19 pandemic. Future studies will focus on expanding the dataset.

The Influence of Online Visual Merchandising and Customized Cross-Selling on Traveloka Mobile Apps Intention to Reuse: With Mediation of Visual Cues and Dynamic Personalization

2025-03-10T12:24:32+00:00

This study examined the impact of Online Visual Merchandising (OVM) and Customized Cross Selling (CCS) on the Intention to Reuse (ItR) Traveloka’s mobile app, mediated by Visual Cues (VC) and Dynamic Personalization (DP). With the increasing competition in the online travel industry and the need for platforms to retain users, understanding the factors that drive repurchase intentions is crucial. A survey of 135 app users aged 25-40 was conducted, and data were analyzed using PLS-SEM. Results indicate that while OVM alone did not significantly affect ItR, CCS had a positive impact. VC and DP significantly mediated the relationships between OVM, CCS, and ItR. The findings suggest that aesthetic elements alone are insufficient for driving repetitive behavior; instead, a strategic integration of visual, personalized, and cross selling strategies is crucial. The study supports nudge theory and offers practical insights for optimizing digital commerce applications to enhance repurchase intentions.

MultiResUNet for COVID-19 Lung Infection Segmentation Based on CT Image

2024-11-15T04:56:42+00:00

Image segmentation plays a crucial role in medical image analysis, facilitating the identification and characterization of various pathologies. During the COVID-19 pandemic, this technique has proven valuable for detecting and assessing the severity of infection. Recent advancements in deep learning, particularly convolutional neural networks (CNNs), have significantly enhanced the efficacy of image segmentation. Numerous CNN-based architectures have been proposed in the literature, with MultiResUNet emerging as a promising approach. This study investigates the application of the MultiResUNet architecture for segmenting regions of COVID-19 infection within patient lung CT images. Experimental results demonstrate the effectiveness of MultiResUNet, achieving an average Dice score of 73.10%.

Optimizing Sentiment Analysis of Electric Vehicles Through Oversampling Techniques on YouTube Comments

2025-02-23T06:42:10+00:00

Air pollution from motorized fuel vehicles causes adverse impacts on the environment and human health, driving the need for more sustainable alternatives such as electric vehicles. However, the transition to electric vehicles is often met with mixed responses from the public, reflected by sentiments that are split between positive and negative. This research investigates such sentiments through analyzing comments on the YouTube platform, which are classified using two algorithms, SVM and Naïve Bayes, and three oversampling techniques: Random Oversampling, SMOTE, and ADASYN. A comparative evaluation is conducted to determine the most effective algorithm and oversampling strategy for handling imbalanced sentiment data, where negative comments dominate. Initial experiments showed that Naïve Bayes with SMOTE achieved the best result among baseline models, with 64% accuracy. However, traditional oversampling methods alone were not sufficient to significantly improve classification quality. To address this, the study proposes a hybrid method that combines Easy Data Augmentation (EDA), specifically Synonym Replacement (SR), with oversampling techniques. The proposed method substantially improved performance. Naïve Bayes combined with SR and SMOTE or Random Oversampling achieved 88% accuracy, with F1-scores of 0.84–0.85 for the positive class. The best result was obtained using SVM with SR and Random Oversampling, reaching 97% accuracy and F1-scores of 0.97 (negative) and 0.96 (positive). These findings demonstrate the effectiveness of combining augmentation and oversampling in improving sentiment classification and provide insights for stakeholders in promoting EV adoption.

Effectiveness of Differentiated Explorative Flipbooks to Improve The Learning Independence of Junior High School Student

2024-10-04T22:37:26+00:00

This research was conducted to overcome the low level of learning independence of students as indicated by dependent behavior among students at the junior high school level. The ADDIE development model was chosen as a model for developing and implementing learning tools to overcome the problem of student learning independence. Data was collected through non-tests using the Learning Object Review Instrument, User Experience Questionnaire and questionnaires to measure students' learning independence. Data analysis was carried out using percentages and n-gain scores. The result of this research was the creation of a product called Mekdi with an average gain score increase of 0.39 which is in the medium category. There are various elements that support the learning process and increase students' learning independence in this media including flipbook elements, GeoGebra exploration, diagnostic assessments, collaboration spaces, and ice breaking activity.

Optimizing Brain Tumor MRI Classification With Transfer Learning: A Performance Comparison of Pre-Trained CNN Models

2025-01-14T01:35:49+00:00

This study aims to classify brain MRI images into several types of brain tumors using the Convolutional Neural Network (CNN) approach with transfer learning. This method has the advantage of processing complex images in a shorter time than conventional CNN approaches. In this study, the data used was a public database from Kaggle, which consisted of four categories: glioma, meningioma, no tumor, and pituitary. Before entering the transfer learning process, data augmentation is carried out on the training data. Four pre-trained CNN models were used: VGG19, ResNet50, InceptionV3, and DenseNet121. The four models compared their ability to classify MRI images with several evaluation metrics: accuracy, precision, recall, and F1 score. The results of the performance comparison of the four pre-trained models show that the ResNet50 is the best model, with an accuracy of 98%. Meanwhile, VGG19, DenseNet121, and InceptionV3 produce 97%, 96%, and 95% accuracy, respectively. The ResNet50 architecture demonstrated superior performance in brain tumor classification, achieving 98% accuracy. It can be attributed to its residual learning structure, which efficiently manages complex MRI features. Further research should concentrate on larger, more diverse datasets and advanced preprocessing techniques to enhance model generalizability.

Enhancing Diesel Backup Power Forecasting With LSTM, GRU, and Autoencoder-based Input Encoding

2025-03-10T02:23:54+00:00

Ensuring a reliable electricity supply is crucial for Indonesia's development. This study applies deep learning to forecast diesel backup power output. One challenge in such predictions is balancing the input sequence length and the number of features to avoid overly long input sequences, which may degrade model performance. To address this, we utilized an autoencoder to compress the input sequence, improving prediction accuracy. Additionally, given the time-consuming nature of hyper-parameter optimization in deep learning, we employed Bayesian optimization to streamline the process and achieve optimal hyper-parameter settings.The study compares a General Regression Neural Network (GRNN) optimized by FOA with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models optimized by Gaussian Process (GP). Results show that LSTM and GRU with encoded inputs outperform their non-encoded counterparts. The GRU, combined with an autoencoder and Bayesian-optimized hyper-parameters, achieves the lowest prediction error, demonstrating superior forecasting capability.The dataset, obtained from evaluated feeders in Kapuas District, Central Kalimantan, covers hourly power generation and distribution from October 2017 to September 2018. Data was split into 11 months for training and 1 month for testing, with the training set further divided into 70% training and 30% validation. The best performing model achieved RMSE and MAE values of 27.5824 and 14.9804, respectively. Future research may explore further optimization, feature selection techniques, and extended dataset variations.

Application of Deep Reinforcement Learning for Stock Trading on The Indonesia Stock Exchange

2025-02-05T01:06:18+00:00

In the last couple of years, stock trading has gained so much popularity because of its promising returns. However, most investors do not pay attention to the risks of trading without analysis, which can lead to a big loss. Some to reduce these risks, try their luck with automated and pre-programmed trading systems, which are called Expert Advisors. The current study examines the application of DRL for automated assistance in trading with an emphasis on decision-making enhancement, particularly the use of DRL in order to realize high asset returns with a low risk of exposure. Concretely, the two applied DRL methods within this work are A2C and PPO. By systematic testing, the A2C method produced a Sharpe Ratio of 1.6009 with a cumulative return of 1.4468, while the PPO method achieved a Sharpe Ratio of 1.7628 with a cumulative return of 1.4767. These were fine-tuned for the most optimal learning rates, cut loss, and take profit ratios, thus showing great promise with the capability to tune up trading strategies and improve trading performances. The research leverages these DRL techniques, hence arriving at better trading strategies that balance profit and risk, while underlining the promise of advanced algorithms in automated stock trading.

Multimodel Prediction Score Based on Academic Procrastination Behavior in E-Learning

2025-01-24T06:15:03+00:00

This research investigates the impact of academic procrastination on student performance in online learning environments and explores a multimodel approach for grade prediction. Academic procrastination is a well-documented issue that negatively affects learning outcomes, often leading to lower academic performance and increased dropout rates in self-paced learning platforms. This study analyzes behavioral data from 377 students, extracted from Moodle activity logs, which record real-time student interactions with learning materials. To address the gap in understanding procrastination patterns through activity logs, key procrastination-related features were derived from timestamps of task access, submission, and engagement duration. Using K-Means clustering with the Elbow method, students were categorized into three procrastination clusters: low procrastination with high academic performance, high procrastination with low performance, and moderate procrastination with average performance. Seven machine learning models were evaluated for predicting student grades, with Random Forest (RF) achieving the highest accuracy (R² = 0.812, MAE = 6.248, RMSE = 8.456). These findings highlight the potential of using activity logs to analyze procrastination patterns and predict student performance, allowing educators to develop early intervention strategies that support at-risk students and improve learning outcomes.