JURNAL INFOTEL

Recursive Feature Elimination Optimization Using Shapley Additive Explanations in Software Defect Prediction with LightGBM Classification

2025-04-17T10:29:19+00:00

Software defect refers to issues where the software does not function properly. The mistakes in the software development process are the reasons for software defects. Software defect prediction is performed to ensure the software is defect-free. Machine learning classification is used to classify defects in software. To improve the classification model, it is necessary to select the best features from the dataset. Recursive Feature Elimination (RFE) is a feature selection method. Shapley Additive Explanations (SHAP) is a method that can optimize feature selection algorithms to produce better results. In this research, the popular boosting algorithm LightGBM will be selected as a classifier to predict software defects. Meanwhile, RFE-SHAP will be used for feature selection to identify the best subset of features. The results and discussion show that RFE-SHAP feature selection slightly outperforms RFE, with average AUC values of 0.864 and 0.858, respectively. Moreover, RFE-SHAP produces more significant results in feature selection compared to RFE. The RFE feature selection T-Test results are P_value = 0.039 < α = 0.05 and t_coun_t = 3.011 > t_table = 2.776. On the contrary, the RFE-SHAP feature selection T-Test results are P_value = 0.000 < α = 0.05 and t_coun_t = 11.91 > t_table = 2.776.

Object Detection of Hospital Assets Using Computer Vision with Generative Adversarial Networks Method

2025-04-08T03:07:07+00:00

Hospital asset monitoring systems encounter significant challenges in managing partially occluded medical equipment, which affects inventory management and operational efficiency. Conventional object detection methods have shown limitations in accurately detecting occluded medical equipment, potentially leading to asset management inefficiencies. This study presents an integrated framework that combines Generative Adversarial Networks (GAN) inpainting with YOLOv8 to improve the detection accuracy of partially occluded medical equipment. The proposed system was evaluated using three distinct training configurations of 500, 750, and 1000 epochs on a comprehensive medical equipment dataset. The experimental results indicate that the 1000-epoch GAN model demonstrated superior reconstruction performance, achieving a Peak Signal-to-Noise Ratio (PSNR) of 39.68 dB, Structural Similarity Index Measure (SSIM) of 0.9910, and Mean Squared Error (MSE) of 7.0030. Furthermore, the integrated YOLOv8-GAN framework maintained robust detection performance with an F1-score of 0.933, comparable to the 0.938 achieved with unoccluded original images. The detection confidence scores exhibited improvement at higher epochs, ranging from 0.824 to 0.861, suggesting enhanced performance with extended training duration. The findings demonstrate that the integration of GAN inpainting with YOLOv8 effectively enhances occluded object detection in hospital environments, offering a viable solution for improved asset monitoring systems.

Designing User Experience Mobile Augmented Reality for Class XI Chemistry Learning with User Centered Design

2025-04-17T03:24:50+00:00

This study addresses the challenges in high school chemistry education, particularly in understanding hydrocarbon compounds, by designing a Mobile Augmented Reality (MAR) application using the User-Centered Design (UCD) method. The research focuses on enhancing the visualization of submicroscopic and symbolic aspects of chemistry, which students often find abstract and complex. Through iterative design processes involving teachers and students, the study developed an interactive MAR application that displays virtual ball-and-stick models of hydrocarbon compounds. The application was evaluated using the User Experience Questionnaire (UEQ), measuring six aspects of user experience: attractiveness, perspicuity, efficiency, dependability, stimulation, and novelty. Results showed significant improvements across all dimensions from the initial to the final iteration, with five out of six scales achieving "Excellent" ratings in the final version. The study demonstrates the effectiveness of UCD in creating an engaging and user-friendly educational tool, highlighting the potential of MAR technology to address longstanding challenges in chemistry education. The positive user perceptions suggest that when designed carefully considering user needs, MAR applications can significantly enhance the chemistry learning experience for high school students.

The Evaluation of Effects of Oversampling and Word Embedding on Sentiment Analysis

2025-04-17T10:12:31+00:00

Generally, opinion datasets for sentiment analysis are in an unbalanced condition. Unbalanced data tends to have a bias in favor of classification in the majority class. Data balancing by adding synthetic data to the minority class requires an oversampling strategy. This research aims to overcome this imbalance by combining oversampling and word embedding (Word2Vec or FastText). We convert the opinion dataset into a sentence vector, and then an oversampling method is applied here. We use 5 (five) datasets from comments on YouTube videos with several differences in terms, number of records, and imbalance conditions. We observed increased sentiment analysis accuracy with combining Word2Vec or FastText with 3 (three) oversampling methods: SMOTE, Borderline SMOTE, or ADASYN. Random Forest is used as machine learning in the classification model, and Confusion Matrix is used for validation. Model performance measurement uses accuracy and F-measure. After testing with five datasets, the performance of the Word2Vec method is almost equal to FastText. Meanwhile, the best oversampling method is Borderline SMOTE. Combining Word2Vec or FastText with Borderline SMOTE could be the best choice because of its accuracy score and F-measure reaching 91.0% - 91.3%. It is hoped that the sentiment analysis model using Word2Vec or FastText with Borderline SMOTE can become a high-performance alternative model.

Performance of SVM Optimized with PSO as Classification Method for Sentiment Analysis UNNES’s Social Media

2025-04-21T04:31:26+00:00

The rapid growth of Big Data, particularly from social media platforms, presents organizations with vast opportunities for extracting valuable insights. For educational institutions like UNNES, sentiment analysis can be crucial for monitoring and enhancing public perception. This research explores the application of sentiment analysis using SVM optimized by PSO to improve classification accuracy. Although SVM is widely known for its effectiveness in linearly separable data, it struggles with nonlinear data. By employing kernel functions and optimizing hyperparameters through PSO, this study aims to improve SVM’s performance. The results show that the optimized SVM model with the RBF kernel and PSO achieved an accuracy of 82.05%, compared to 80.96% using standard SVM, demonstrating a 1.09% improvement. These findings indicate that PSO significantly enhances the efficiency and accuracy of SVM models in sentiment analysis, making it a powerful tool for analyzing social media data in educational contexts.

Brain Tumor Detection Through Image Enhancement Methods and Transfer Learning Techniques

2025-04-23T07:03:39+00:00

A brain tumor is dangerous and must be treated immediately to prevent worsening. The identification of brain tumors can be performed by a more in-depth examination by specialists or by using artificial intelligence technology through MRI datasets. Several studies have examined how artificial intelligence could be used to find brain cancer in MRI images. The algorithm usually used is CNN with the addition of transfer learning. Previous studies have produced very high accuracy, but the accuracy value can still be improved. In this study, MRI image quality is improved as a new input for modeling. The test results show that the proposed CNN Model produces an accuracy of 98.50% on the test data. This result is higher than the baseline method of 98.45%. Analysis of other metrics, such as precision, recall, and F1-score, indicates consistent performance across classes. These findings suggest that using preprocessing to improve image quality can improve Model performance. Using CLAHE and median blur to improve image quality can improve accuracy by 14.5%. This study contributes to identifying an effective combination of Model optimization techniques for image classification tasks.

Improving the Accuracy of Concrete Mix Type Recognition with ANN and GLCM Features Based on Image Resolution

2025-04-23T07:09:02+00:00

Concrete is an essential construction material that is often used due to its strength and durability, but its mix type identification often relies on conventional methods that are less efficient and accurate. This research aims to evaluate the effect of image resolution on the accuracy of concrete mix type recognition using Artificial Neural Network (ANN) and Gray-Level Co-Occurrence Matrix (GLCM) features. The method used involves analysing concrete images at various resolutions: 200 x 200, 300 x 300, 400 x 400, 500 x 500, 600 x 600, and 700 x 700 pixels. The experimental results show that higher image resolutions tend to improve recognition accuracy. all types of image sizes using 1,250 training data and 250 test data. Image sizes of 200 x 200 and 300 x 300 pixels give low accuracy of 42% and 45% respectively, while sizes of 400 x 400 and 500 x 500 pixels show an increase in accuracy to 60.5% and 62.5%. The higher resolutions of 600 x 600 and 700 x 700 pixels produced the highest accuracy of 68% and 70%, respectively. These results indicate that larger image resolutions are able to capture more details and characteristics required for more accurate concrete mix type recognition. This research has implications for improving efficiency and consistency in concrete inspection in the construction industry through the use of AI-based image recognition methods.