Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Not a member yet
1071 research outputs found
Sort by
Comparative Analysis of Machine Learning Algorithms for Predicting Patient Admission in Emergency Departments Using EHR Data
No full textEvery patient who is rushed to the Emergency Department needs fast treatment to determine whether the patient should be inpatient or outpatient. However, the existing fact is that deciding whether an inpatient or outpatient must wait for the diagnosis made by the existing doctor, so if there are many patients, it generally takes quite a long time. So, to predict patient admissions to the emergency unit, a machine learning model that can be fast and accurate is needed. Therefore, this study developed a machine learning and neural network model to determine patient care in Emergency Departments. This study uses publicly available electronic health record (EHR) data, which is 3,309. The model development process uses machine learning methods (SVM, Decision Tree, KNN, AdaBoost, MLPClassifier) and neural networks. The model that has been obtained is then evaluated for its performance using a confusion matrix and several matrices such as accuracy, precision, recall, and F1-Score. The results of the model performance evaluation were compared, and the best model was obtained, namely the MLPClassifier model with an accuracy value = 0.736 and an F1-Score value = 0.635, and the Neural Network model obtained an accuracy value = 0.724 and an F1-Score value = 0.640. The best models obtained in this study, namely the MLPClassifier and Neural Network models, were proven to be able to outperform other models
Optimizing Sensitivity in Machine Learning Models for Pediatric Post-operative Kyphosis Prediction
Get PDFPost-operative kyphosis represents a significant complication following pediatric spinal corrective surgery, necessitating sophisticated prediction methods to identify high-risk patients. This study developed and evaluated machine learning models for kyphosis prediction using a dataset of 81 pediatric patients by comparing the logistic regression and decision tree approaches. Despite achieving a higher overall accuracy (82%), the logistic regression model failed to identify any kyphosis cases, rendering it clinically ineffective. Conversely, the decision tree model demonstrated superior clinical utility by successfully identifying 33% of kyphosis cases while maintaining 71% accuracy. Feature importance analysis established starting vertebral position as the dominant predictor (importance=0.554), followed by patient age (0.416), with vertebrae count contributing minimally (0.030). The decision tree identified critical thresholds for risk stratification: operations beginning at or above T8-T9, particularly in children aged 5-9 years, carried a substantially elevated kyphosis risk. Our methodological approach emphasizes sensitivity over conventional accuracy metrics, recognizing that missing high-risk patients have greater clinical consequences than unnecessary monitoring. This study demonstrates the capacity of decision tree models to extract clinically meaningful patterns from small, imbalanced surgical datasets that elude conventional statistical approaches
University Students Stress Detection During Final Report Subject by Using NASA TLX Method and Logistic Regression
Get PDFStress is a psychological response that occurs when someone faces pressure or demands that exceed their ability to adapt. In the context of a final-year student, stress is often a significant problem due to academic pressure, such as completing final assignments, as well as demands to immediately prepare to enter the workforce and demands to immediately prepare to enter the workforce. Research shows that stress that is not managed properly can cause various negative effects, such as sleep disorders and decreased cognitive function. This study aimed to identify and analyze stress levels among final-year students who completed a final report by integrating physiological and psychological data. In this study, 30 students were assessed using a wearable system to obtain physiological data, such as heart rate and body temperature, while subjective assessments were carried out using the NASA-TLX method to measure mental workload. The results showed that 19 out of 30 respondents experienced significant levels of stress and 11 respondents were in normal conditions, with the main causal factors including high academic pressure and distance regarding the future. In addition, the logistic regression analysis applied in this study succeeded in developing a predictive model with an accuracy of 94% in identifying students' stress conditions. This shows that this method is sufficiently accurate for detecting stress symptoms in final-year students
Automatic Classification of Multilanguage Scientific Papers to the Sustainable Development Goals Using Transfer Learning
Get PDFThe classification of scientific papers according to their relevance to Sustainable Development Goals (SDGs) is a critical task in identifying the research development status of goals. However, with the growing volume of scientific literature published worldwide in multiple languages, manual categorization of these papers has become increasingly complex and time-consuming. Furthermore, the need for a comprehensive multilingual dataset to train effective models complicates the task, as obtaining such datasets for various languages is resource intensive. This study proposes a solution to this problem by leveraging transfer learning techniques to automatically classify scientific papers into SDG labels. By fine-tuning pretrained multilingual models mBERT on SDG publication datasets in a multilabel approach, we demonstrate that transfer learning can significantly improve classification performance, even with limited labelled data, compared to SVM. Our approach enables the effective processing of scientific papers in different languages and facilitates the seamless mapping of research to the relevance of SDGs, the four pillars of SDGs, and the 17 goals of SDGs. The proposed method addresses the scalability issue in SDG classification and lays the groundwork for more efficient systems that can handle the multilingual nature of modern scientific publications
Application of YOLOv8 Algorithm for Coral Reef Disease Detection as an Effort to Prevent Marine Habitat Damage in Batam
No full textResearch in 2019 in Batam City showed that out of 19 coral reef fisheries support facilities, 16 were declared not good. Coral reef damage increased from 36.28% to 39.44%. This is due to the threat of coral reef damage due to international shipping lane areas, human activities such as destructive fishing, pollution, sedimentation, and global warming. These threats can cause coral diseases such as black band disease (BBD), brown band disease (BrB), Bleaching Coral, and yellow band disease (YBD). The Underwater Photo Transect (UPT) method collects data in the field in the form of underwater photos and analyzes them to obtain quantitative data. This method has a weakness, namely the low level of accuracy in detecting coral reef diseases. This study proposes coral reef disease detection using the YOLO model YOLO8l, YOLO8x, and YOLO8m. The results of the model evaluation test with a threshold value of 0.5 to 0.95 against the test data show that the three models can detect coral reef diseases with an accuracy of 99%. These results prove that the YOLOv8 model in this study is suitable for the real-time detection of coral reef diseases to replace the Underwater Photo Transect (UPT) method, which has low accuracy. Applying the YOLOv8 method will help Prevent Marine Habitat Damage in Batam City
Enhancing Lung Cancer Detection: Optimizing CNN Architectures through Hyperparameter Tuning
No full textTThis study aimed to compare the performance of various Convolutional Neural Network (CNN) architectures, including LeNet, ResNet, AlexNet, GoogleNet, VGGNet, and the proposed model, in medical image classification for disease detection. The proposed model was developed by adding additional layers and fine-tuning the hyperparameters in the ResNet architecture to enhance its ability to extract complex features. The training and testing processes were conducted using an augmented X-ray image dataset to increase the data diversity. The results indicate that the proposed model achieved the highest testing accuracy of 76.33%, surpassing other models in terms of accuracy, precision, recall, and F1-score. Although there are some limitations in specificity and the Matthews Correlation Coefficient (MCC), the proposed model still demonstrates better generalization ability, with an AUC-ROC score approaching an optimal value. These findings suggest that the proposed model has advantages in medical image classification and holds potential for further development to enhance disease classification accuracy
Enhanced Predictive Modeling for Non-Invasive Liver Disease Diagnosis
No full textLiver diseases (e.g. cirrhosis, hepatitis, and fatty liver disease) are globally one of the leading causes of mortality and are typically diagnosed in advanced stages due to vague symptoms and the difficulty involved in existing diagnostic techniques (e.g. biopsies). To optimize the early diagnosis of liver disease, this study proposes an enhanced, non-invasive approach using machine learning techniques. The research is enriched with a full pipeline, from exploratory data analysis and imputation of the dataset, treatment of the outlier, encoding of labels and scaling using ILPD (Indian Liver Patient Dataset). The classification models compared were RandomForest, XGBoost, LGBM, and CatBoost. The CatBoost algorithm fine-tuned with RandomizedSearchCV showed the highest performance with a test accuracy of 93%. The performance was again better than any already published methods showing that advanced ensembling and hyperparameter optimization worked. The proposed model is suitable for incorporation into clinical decision support systems and provides reliable and accurate diagnostic assistance. In addition to its high accuracy, the model is robust for missing and categorical data, which is a challenge in any real-world clinical scenario. These findings add to the growing body of evidence supporting AI-based medical diagnostics and suggest that CatBoost is a highly promising tool for facilitating timely screening and diagnosis of liver disease. Furthermore, the study stresses the need for thorough preprocessing and cross-validation, which serve to reduce biases that are present in widely applied datasets. Ongoing future efforts may involve the integration of multi-source data and implementation of explainable AI techniques to allow for wider clinical trust and use
Optimizing Tourism Recommendations with a Hybrid Model: Bridging User Preferences and Behavioral Patterns
No full textRecommender systems play a crucial role in personalized decision-making, particularly in the tourism industry, where users seek destinations that align with their preferences. However, traditional recommendation methods often struggle to provide accurate recommendations. This study proposes a hybrid recommendation model that integrates Content-Based Filtering (CBF) and Apriori association rule mining to enhance recommendation quality. First, CBF was implemented using TF-IDF, Word2Vec, and BERT embeddings to compute the similarity between user preferences and tourism destinations. The Top-N recommended destinations from each method were then used as antecedents in Apriori to identify associative patterns and co-occurrence relationships among tourism destinations. By leveraging both semantic preference matching and association rule mining, the proposed system refines the recommendation process, ensuring not only personalized suggestions but also uncovering implicit travel patterns. The experimental results demonstrate that the hybrid model improves recommendation relevance and accuracy compared to standalone CBF methods. The accuracy of the CBF model was 53.96%, whereas that of the hybrid model was 94.31%. The integration of CBF and Apriori offers a more comprehensive and data-driven recommendation framework, which is valuable for personalized tourism applications
Optimizing DBSCAN Parameters for Depth-Based Earthquake Clustering Using Grid Search
No full textThis study addresses the challenge of accurately clustering earthquake events based on depth to better understand seismic activity patterns in Sulawesi from 2019 to 2023. Traditional clustering algorithms often fail to capture the complex spatial and depth-based structures of earthquake data. To overcome this, we employed the DBSCAN algorithm, which is well-suited for identifying irregularly shaped clusters and handling noise in spatial datasets. A key focus of this research is the systematic optimization of DBSCAN’s parameters—epsilon (ε) and minimum samples (min_samples)—using a grid search approach. Epsilon values varied from 0.1 to 0.5, and min_samples ranged from 6 to 60. The optimal parameters, determined using the Calinski-Harabasz (CH) index, were ε = 0.4 and min_samples = 54. Compared with previous heuristic settings, the optimized configuration produced better separated and more interpretable clusters. Using the optimized parameters, nine distinct clusters were identified, capturing meaningful patterns in both depth and magnitude. The results revealed that shallow earthquakes (0–20 km) tend to exhibit greater magnitude variation, with some clusters averaging magnitudes up to 3.7. This suggests a higher seismic hazard potential associated with brittle crustal activity. The findings contribute to seismic hazard analysis by providing a more robust understanding of three-dimensional earthquake distribution, aiding regional risk assessment and disaster preparedness efforts. These insights can support agencies such as BMKG and BPBD in hazard mapping, sensor deployment, and contingency planning for high-risk zones
Thermal Comfort Projection on Northern Coast of Central Java Using Machine Learning
No full textThe Thermal Humidity Index (THI) serves as a critical measure of environmental thermal comfort, particularly vital for living beings in densely populated regions. This study projects and classifies THI in the western northern coastal areas of Central Java using Machine Learning (ML) techniques. Utilizing temperature and humidity data from 2018 to 2024, THI projections were conducted using the XGBoost algorithm, whereas comfort level classifications were performed using the Random Forest algorithm. The results indicate that Semarang City, eastern Kendal, Pemalang, and Tegal frequently experienced slightly uncomfortable conditions (THI 27–30), particularly during the rainy and transitional seasons, whereas other regions maintained comfortable levels (THI < 27). The THI projection model for 2025–2029 achieved an accuracy of 73%, while the classification model attained a remarkably high accuracy of 99.94%. These findings highlight the need for enhanced regional management strategies in areas with reduced thermal comfort