Matrix : Jurnal Manajemen Teknologi dan Informatika

Application of conditional lightweight GAN for retinal fundus image synthesis based on diabetic retinopathy severity levels on the IDRiD dataset

2026-02-19T04:54:31+00:00

Diabetic Retinopathy (DR) is a leading cause of preventable blindness, yet the development of automated diagnostic models using Deep Learning is often hindered by the availability of imbalanced medical datasets. This study aims to address this issue by implementing a Conditional Lightweight Generative Adversarial Network (c-LGAN) architecture to synthesize realistic fundus retinal images corresponding to five DR severity levels from the IDRiD dataset. The c-LGAN model was trained on a balanced dataset, and its performance was quantitatively evaluated using Frechet Inception Distance (FID) and Inception Score (IS) metrics. The results demonstrate that the proposed model is capable of generating high-quality images, evidenced by achieving a best FID score of 121.24 at epoch 100. However, further observation identified significant stability challenges in long-term training, marked by a performance collapse after the model reached its optimal point. This phenomenon was attributed to an overpowering discriminator. This study concludes that c-LGAN is a promising approach for data augmentation but emphasizes the critical importance of periodic metric monitoring and model checkpointing strategies to capture peak performance and overcome training stability issues.

Integration of local wisdom and modern medicine in a treatment recommendation system for toddlers based on the Case-Based Reasoning-Fuzzy Method

2025-12-09T01:19:54+00:00

This study aims to develop a treatment recommendation system for toddlers based on Case-Based Reasoning (CBR) combined with Fuzzy Logic, by integrating modern medical knowledge and local wisdom. The system was developed to address the need for adaptive initial diagnosis recommendations, particularly in addressing ambiguous symptoms. At the case representation stage, disease, symptom, and drug data from medical and traditional perspectives are used as the knowledge base. The CBR process serves as the primary mechanism for searching for similar cases, while fuzzy logic is used at the revision stage to provide degrees of symptom intensity so that the diagnosis results are more flexible. System evaluation was conducted through blackbox testing, accuracy measurements, and the System Usability Scale (SUS) method involving 50 respondents. The results showed that all system functions ran as planned, the accuracy level reached 88%, and the average SUS score was 78.4 in the Good Usability category, indicating the system is easy to use and user-acceptable. This study proves that the CBR–Fuzzy integration is effective in providing accurate, adaptive, and culturally relevant initial diagnosis recommendations. For further research, it is recommended to expand the case base, refine fuzzy rules, develop a broader interface, and implement the system on a mobile platform to improve accuracy, ease of access, and wider user acceptance.

Bell’s Palsy and stroke face classification using SVM with MediaPipe Face Mesh

2026-02-14T11:36:56+00:00

Stroke and Bell’s Palsy share similar manifestations of unilateral facial paralysis, often leading to clinical misinterpretation, particularly in acute cases. Although deep learning approaches have demonstrated strong performance in segmenting facial paralysis regions, these methods primarily focus on area localization rather than directly differentiating Stroke and Bell’s Palsy, and typically require large-scale datasets and substantial computational resources. To address this gap, this study proposes an explainable and resource-efficient framework for classifying Stroke and Bell’s Palsy using asymmetric facial numeric features extracted from static images. Unlike appearance-based deep learning models, the proposed approach transforms facial landmarks detected by MediaPipe Face Mesh into geometric asymmetry features through Min–Max scaling, Euclidean distance, and angle computation. After class balancing via undersampling, classification was performed using an SVM with an RBF kernel. The 70:30 split achieved the most stable performance, with a testing accuracy of 0.8041 and cross-validation accuracy of 0.8072 ± 0.0069, indicating minimal generalization gap. These findings demonstrate that geometric asymmetry features combined with SVM provide a reliable and interpretable alternative for differentiating BP and ST under limited data and computational constraints.

User experience testing on Smart Human Capital Dashboard (SHUCADA) from PT Studio Kami Mandiri using User Experience Questionnaire (UEQ)

2026-02-19T12:39:43+00:00

PT Studio Kami Mandiri is a company specialized in software development. PT Studio Kami Mandiri has developed several products or applications that have also been used by large companies. One of the products developed is the Smart Human Capital Dashboard (SHUCADA) which is used to monitor the performance of an employee in a company. User experience testing is needed to find out the possible obstacles that users will face when using the application. Therefore, this research conducted user experience testing on SHUCADA to find out employee perceptions of the application. User Experience Testing of SHUCADA using the User Experience Questionnaire (UEQ) Data Analysis Tool found that aspects of attractiveness with mean value 1.21, dependability with mean value 1.208, and stimulation with mean value 1.125 get above average scores, and aspects of efficiency (mean 1.583) get good scores. The perspicuity aspect (mean 0.604) received a bad score and the novelty aspect (mean 0.604) received a below average score.

Model creation for Denial of Service (DoS) attack classification using an ensemble learning approach on multi-dataset network traffic

2026-02-06T01:47:19+00:00

The rapid advancement of information technology has increased cybersecurity threats, one of which is the Denial of Service (DoS) attack that can disrupt service availability. Most existing studies on DoS attack classification rely on a single dataset and a single machine learning model, which limits the generalizability of their results across different network environments. This study addresses this gap by proposing an ensemble learning-based model for DoS attack classification using multi-dataset network traffic. The datasets used in this research are UNSW-NB15 and TON-IoT, which were combined based on feature compatibility. After the preprocessing stage, a final dataset consisting of 73,302 records was obtained, comprising 64,267 normal traffic instances and 9,035 DoS attack instances. The dataset was then split using stratified sampling with an 80:20 ratio for training and testing data. The ensemble learning methods applied include Random Forest (bagging) and XGBoost (boosting), with training scenarios using both the original dataset and data balanced using the Synthetic Minority Over-sampling Technique (SMOTE). Model evaluation was conducted using a confusion matrix and performance metrics including accuracy, precision, recall, F1-score, and ROC-AUC.The results show that the ensemble learning approach achieves high performance in classifying DoS attacks. However, the application of SMOTE did not improve model performance in this study. The best-performing model was Random Forest trained on the original dataset, achieving an accuracy of 0.9854, precision of 0.9515, recall of 0.928, F1-score of 0.9402, and ROC-AUC of 0.996. These results indicate that the proposed model is effective for DoS attack classification across heterogeneous network traffic data.