IJCCS (Indonesian Journal of Computing and Cybernetics Systems)
Not a member yet
480 research outputs found
Sort by
Predicting Price and Risk ICBP Stocks Using GRU and VaR
Full text linkThe economy plays a vital role in maintaining a country’s stability and progress, where stock investments serve as a primary financial instrument to enhance societal welfare. In Indonesia, interest in stock investments, especially in the essential food sector, continues to grow due to its long-term profit potential. This study combines stock price prediction with risk analysis using a Gated Recurrent Unit (GRU) model and Value at Risk (VaR) calculation based on historical simulation. The GRU model is selected for stock price prediction due to its ability to capture complex, fluctuating patterns and adapt to market changes, while VaR is used to measure potential maximum loss at a 95% confidence level. The findings indicate a potential loss of IDR 65.785, demonstrating that this approach can provide a risk estimate by combining future predicted prices with historical data. Thus, this approach offers guidance for investors in understanding potential profits and risks in stock assets. The integration of GRU-based predictions and historical simulation VaR is expected to support more informative and prudent investment decision-making, particularly in facing the dynamic and risky stock market conditions
Comparison of Artificial Intelligence Methods for Tuberculosis Detection Using X-Ray Images
Full text linkPenyakit tuberkulosis (TB), yang disebabkan oleh bakteri Mycobacterium tuberculosis, merupakan penyakit menular yang sangat berbahaya. Di Indonesia, TB adalah penyakit menular paling mematikan setelah COVID-19 dan menempati urutan ke-13 sebagai penyebab kematian global. Deteksi dini TB sangat penting untuk meningkatkan peluang kesembuhan, namun keterbatasan jumlah ahli radiologi menjadi tantangan utama. Teknologi deep learning, khususnya Convolutional Neural Network (CNN), mejadi solusi efektif untuk masalah ini. Oleh karena itu, pada penelitian ini akan membandingkan dua arsitektur CNN, yaitu AlexNet dan VGG-19, dalam mendeteksi TB pada citra rontgen paru-paru, dengan penerapan metode perbaikan kualitas citra, seperti Histogram Equalization (HE), Adaptive Histogram Equalization (AHE), Contrast Limited Adaptive Histogram Equalization (CLAHE), dan Gamma Correction. Dataset yang digunakan diperoleh dari Kaggle dan mencakup citra rontgen paru-paru normal serta TB. Evaluasi performa dilakukan berdasarkan akurasi, presisi, recall, dan F1-score. Hasil penelitian menunjukkan bahwa VGG-19 dengan CLAHE memberikan performa terbaik dengan akurasi 93.5%, presisi 98.88%, recall 88%, dan F1-score 93.12%. VGG-19 dengan Gamma Correction juga menunjukkan hasil yang sangat baik dengan akurasi 91%, presisi 97.67%, recall 84%, dan F1-score 90.32%. Temuan ini menggarisbawahi efektivitas kombinasi CNN dan metode pemrosesan citra dalam meningkatkan deteksi TB
Optimization of Palm Fruit Ripeness Detection With Yolov11 on CPU
Full text linkThe palm oil industry is one of the strategic sectors that contributes significantly to the Indonesian economy. However, this industry still faces various challenges, particularly in terms of operational efficiency and the implementation of digitalization, especially at the level of independent farmers who often still use manual methods to determine the ripeness of the fruit. This manual process is prone to subjectivity, which can impact harvest quality and supply chain efficiency. To address this issue, this study proposes a palm oil fruit ripeness detection system based on the YOLOv11 algorithm, chosen for its advantages in inference speed and detection accuracy, especially when run on devices with limited resources. The developed model was then implemented using the ONNX Runtime Framework. This enables accelerated inference processes and supports portability on hardware with limited resources. Test results show that the model achieves an mAP@50 accuracy of 90.2% with an average latency of around 255 ms to 300 ms. With these achievements, this system is not only reliable in detecting fruit ripeness, but also efficient in processing time and relevant to support digital transformation in the palm oil plantation sector
Optimization of Gradient Boosting Method for Predicting Narcissistic Personality Disorder (NPD) in Employees
Full text linkNarcissistic Personality Disorder (NPD) is a serious challenge in modern workplace environments; however, early detection and appropriate intervention remain unmet needs. This research aims to address the issue by proposing an intelligent system model based on machine learning, utilizing the Gradient Boosting method to predict NPD. The Gradient Boosting method was chosen for its ability to handle complex data and gradually improve prediction performance. This model is integrated with employee data, including a range of psychological, behavioral, and demographic variables relevant to NPD. The primary contribution of this research is the development of a predictive model that can assist organizations in identifying and providing early intervention to employees at risk of developing NPD. In doing so, it is expected to reduce the negative impact of NPD on the workplace, such as interpersonal conflicts and decreased productivity. The study shows significant results in the model's classification performance after applying Recursive Feature Elimination (RFE) to optimize the Gradient Boosting method. The accuracy rate reached 82%, an improvement from the previous 79% achieved using the Gradient Boosting Classifier. This indicates that the RFE-Gradient Boosting model has greater potential in classifying employees who genuinely have narcissistic personality disorder versus those who do not
Prediction Sentiment Analysis Grab Reviews Using SVM Linear Based Streamlit
Full text linkAdvances in digital technology have accelerated the transformation of online transportation services, intensifying competition and driving innovations to enhance service quality. As a leading platform in Indonesia, Grab faces various challenges, including driver service quality, payment systems, and application stability, as reflected in user reviews on Google Play Store. This study aims to gain strategic insights by evaluating a linear kernel-based Support Vector Machine (SVM) model integrated into the Streamlit platform to predict the sentiment of Grab user reviews. Data were collected via web scraping and processed using tokenization, stopword removal, and stemming techniques to improve model accuracy. The model was implemented on an interactive Streamlit website featuring two main functionalities: sentiment prediction and plot visualization. The sentiment prediction feature presents sentiment distribution, performance metrics, a confusion matrix, and a classification report, while the visualization feature displays interactive word clouds, bar charts, and pie charts. Model evaluation reveals an accuracy of 83% in the Streamlit environment. These findings are expected to contribute to developers and stakeholders in enhancing Grab services and advancing more effective sentiment prediction methods
Classifying Indonesian Hoax News Titles with SVM, XGBoost, and BiLSTM
Full text linkThis study investigates the automated detection of hoaxes related to President Jokowi in Indonesian news by analyzing only news titles, aiming for efficient detection and reduced traffic to harmful websites. We compared the performance of traditional (SVM, XGBoost) and deep learning (BiLSTM) algorithms, with and without Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance in a dataset scraped from trusted news sources (CNN Indonesia, Detik News) and a fact-checking platform (turnbackhoax.id). The results indicate that BiLSTM generally outperformed SVM and XGBoost, demonstrating the potential of deep learning for this task. However, applying SMOTE negatively impacted BiLSTM's performance, suggesting overfitting. Notably, precision consistently exceeded recall across all models, indicating high reliability in identifying hoaxes but a potential for missing a significant number of actual hoaxes. This highlights a trade-off between avoiding false positives and ensuring comprehensive detection. The findings also suggest that language-specific characteristics influence algorithm effectiveness. This research contributes to developing efficient and accurate tools for combating misinformation in the Indonesian online environment, emphasizing the importance of title-based analysis and careful consideration on data balancing
Court Decision Prediction Model Using Natural Language Processing and Random Forest
Full text linkThe increasing number of criminal cases in Indonesia, which reached 288,472 in 2023, or rose by 15% from the previous year, has created a substantial workload for judicial professionals. This situation highlights the urgent need for artificial intelligence–based decision support systems to accelerate and improve the quality of legal decision-making. This study proposes a court decision prediction approach using the Random Forest algorithm combined with Natural Language Processing (NLP) techniques. The dataset consists of 21,630 court decisions from the Supreme Court of Indonesia, originally in PDF format and converted into XML. The research procedure includes text preprocessing, feature construction using Word2Vec and Fast Text, and Random Forest classification. Unlike previous studies employing LSTM, BiLSTM, and CNN methods with accuracy ranging from 49.14% to 77.32%, the proposed approach delivers better performance. Experimental results show that the model achieves a prediction accuracy of up to 63%-81% for Penalty Categories classification and up to 65%-80% for long punishment regression. These findings demonstrate the significant potential of applying NLP and Random Forest to develop predictive systems in Indonesian legal document analysis
Deep Learning for Automatic Assessment and Feedback in LMS-Based Education
Full text linkLearning Management Systems (LMS) play a critical role in modern education by organizing content, facilitating communication, and supporting student assessment. However, most current LMS platforms depend on manual grading and generalized feedback, which can be inefficient and lack personalization. This research enhances LMS capabilities by integrating deep learning techniques—specifically Natural Language Processing (NLP)—to automate assessment and deliver personalized feedback. The system analyzes student input, such as written assignments and discussion forum posts, to evaluate performance and generate real-time, adaptive feedback. A modular framework was developed using a Bidirectional LSTM-based architecture trained on sequence data with regression objectives. The model was evaluated using the Mean Squared Error (MSE) metric. The results show that the model performs reasonably well, with predictions closely aligned to actual values in most cases, although its performance decreases slightly at the distribution extremes. Visualization via scatter plots further confirms the model's ability to capture context and structure in textual input. These findings demonstrate the model's feasibility in educational environments and its potential to reduce instructor workload while improving the quality of feedback. Future work will consider integrating attention mechanisms and multilingual capabilities for broader applicability
A Comparative Deep Learning Approach for Classifying Oil Palm Fruit Ripeness Levels Using YOLOv8s and Faster R-CNN
Full text linkAssessing oil palm fruit ripeness is essential for optimizing harvest timing and maximizing market value. In many developing regions, harvesting is still performed every 10–15 days through manual visual inspection, a process prone to human error that often causes premature harvesting and reduces selling value by up to 50%. This study explores deep learning-based object detection for automatic classification of oil palm fruit bunches. A dataset of 4,578 annotated high-resolution images was prepared and categorized into six ripeness classes: Empty, Immature, Underripe, Abnormal, Ripe, and Overripe. Two advanced detection models, YOLOv8s and Faster R-CNN with a ResNet-50 backbone, were evaluated under identical conditions using precision, recall, and mean Average Precision (mAP) metrics. YOLOv8s achieved precision and recall above 99%, with a mAP 0.5:0.95 of 0.9254, demonstrating strong reliability and efficiency for real-time use. Faster R-CNN achieved a higher mAP 0.5 of 0.9964, indicating superior localization accuracy but slower computation. Overall, YOLOv8s provides a better trade-off between accuracy and speed, making it more practical for automated harvesting. This research supports precision agriculture by emphasizing AI-driven solutions that improve productivity, minimize losses, and promote sustainable palm oil management
LOOK ALIKE-SOUND ALIKE PREDICTION AS A TOOL FOR PATIENT SAFETY
Full text linkReport from the WHO that one of the highest causes of medication errors is Look Alike – Sound Alike (LASA) drugs, leading to errors in receiving information about the drugs, which of course will affect patient safety. Efforts to reduce medication errors have been widely implemented, such as conducting medication training, managing medications, and storing and labeling medications. However, all of that leads to human error, so the utilization of technology is needed to address this issue. The technology expected to help reduce medication errors is the utilization of artificial intelligence (AI). AI is designed for automation processes and systems that can learn independently, allowing the causes of medication errors such as LASA to be learned by the system and predicted automatically. Deep learning is a part of AI that works by providing solutions accurately and automatically. The Recurrent Neural Networks (RNN) algorithm is one of the deep learning methods that has been proven accurate in predictions based on previous research studies. In this study, LASA predictions were made using RNN with the aim of serving as an aid to reduce medication errors, thereby ensuring patient safety. The accuracy achieved is 99% for training and 81% for testing