Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
UAV Image Classification of Oil Palm Plants Using CNN Ensemble Model
Get PDFBasal Stem Rot (BSR), caused by Ganoderma boninense, is one of the most destructive diseases affecting oil palm plantations in Southeast Asia. Early detection of this disease is crucial to prevent its widespread transmission and to maintain plantation productivity. This study proposes an image classification approach using ensemble learning with three Convolutional Neural Network (CNN) architectures: DenseNet161, ResNet152, and VGG19, to detect BSR-infected oil palm trees through aerial imagery captured by Unmanned Aerial Vehicles (UAVs). The dataset used consists of 7,348 annotated images classified into two categories: healthy and unhealthy. Experimental results show that the DenseNet161 model outperformed the others, achieving a validation accuracy of 91.75% and a validation loss of 0.0307. The ensemble CNN approach demonstrated improved classification accuracy and holds significant potential for implementation in automated and precise plant health monitoring systems. This research provides a valuable contribution to AI-based agricultural technology, particularly in disease management for oil palm plantations.Basal Stem Rot (BSR), caused by Ganoderma boninense, is one of the most destructive diseases affecting oil palm plantations in Southeast Asia. Early detection of this disease is crucial to prevent its widespread transmission and to maintain plantation productivity. This study proposes an image classification approach using ensemble learning with three Convolutional Neural Network (CNN) architectures: DenseNet161, ResNet152, and VGG19, to detect BSR-infected oil palm trees through aerial imagery captured by Unmanned Aerial Vehicles (UAVs). The dataset used consists of 7,348 annotated images classified into two categories: healthy and unhealthy. Experimental results show that the DenseNet161 model outperformed the others, achieving a validation accuracy of 91.75% and a validation loss of 0.0307. The ensemble CNN approach demonstrated improved classification accuracy and holds significant potential for implementation in automated and precise plant health monitoring systems. This research provides a valuable contribution to AI-based agricultural technology, particularly in disease management for oil palm plantations
Visual Segmentation and Classification of Coffee Beans After Roasting
Get PDFThis research aims to develop an image-based system for segmenting and classifying coffee beans after roasting using deep learning. A U-Net architecture was applied to isolate coffee beans from the background with high spatial accuracy, achieving a mean Intersection over Union (IoU) of 0.8833 and Dice Coefficient of 0.9375. The segmented images were then classified into six roasting levels green, light, light to medium, medium, medium to dark, and dark using a modified ResNet-50 model, which reached an overall classification accuracy of 86%. The system demonstrates strong performance for clear categories but shows overlapping predictions for visually similar classes such as “medium” and its neighboring levels, indicating that boundaries between roasting stages can be ambiguous. This study provides an objective and automated alternative for roast quality inspection, reducing reliance on subjective human assessment. However, to meet industrial standards, further improvements are needed, such as integrating additional image features or ensemble models to increase discrimination power. This two-stage system serves as a promising foundation for future developments in automated coffee quality control
Heart Disease Classification Using Extreme Learning Machine (ELM) Method With Outlier Handling One-Class Support Vector Machine (OCSVM)
Get PDFHeart disease remains the leading cause of death globally, accounting for approximately 32% of all deaths. Developing countries are particularly affected due to prevalent risk factors such as hypertension, diabetes, and poor lifestyle habits. Accurate and early diagnosis is essential for effective treatment and prevention. Technological advancements have enabled the precise analysis of complex clinical data. This study investigates the application of the Extreme Learning Machine (ELM) algorithm combined with outlier handling using One-Class Support Vector Machine (OCSVM) for heart disease classification. The dataset, obtained from the University of California, Irvine Machine Learning Repository, consists of 1190 clinical records with 12 numerical features. The ELM model was evaluated using the Tanh activation function and 10-fold cross-validation. Among the tested configurations, the best performance was achieved using 450 hidden neurons, yielding a sensitivity of 92,52% with a standard deviation of 4,00%. These results indicate that ELM, when paired with effective outlier handling and properly tuned parameters, can provide reliable and stable performance in heart disease classification
Analyzing and Controlling COVID-19 Using SageMath Toolbox: A case Study in the D.R. Congo
Get PDFUnderstanding the dynamics of an epidemic, to control, manage, or eradicate it, requires a wealth of knowledge in biology and mathematics. Computer tools also make significant contributions, thus, enabling us to carry out analyses and find approximate solutions, as well as run simulations to determine trends over time. In this study, we present a compartmental SVEIHAR model for the propagation and prevention of COVID-19. Using the computational and mathematical competencies of SageMath software (version 9.3) we simulate and evaluate the spread of the virus. Equilibria are calculated and adjusted according to the data. Again, the basic reproduction number, stabilities, and parameter sensitivities were studied. Our findings indicate that vaccination and cure rates are the most sensitive parameters, playing a crucial role in the fight against COVID-19. Again, the use of traditional plants, prayer, and meditation significantly decreases the value of the basic reproduction number. We also found that the disease will disappear after a time. Lastly, our study has shown the usefulness of SageMath software (version 9.3) which could be adapted to a variety of mathematical epidemic models
Security Risk Analysis of QRIS Implementation in Public Locations Using ISO 31000:2018 Framework
Get PDFThis study aims to analyze the security risks associated with the implementation of the QRIS (Quick Response Indonesia Standard) payment system in public spaces and provide appropriate mitigation recommendations. The research employs a case study approach with a qualitative research design to explore the perceptions of users and business owners regarding the potential risks involved. Data were collected through semi-structured interviews, risk perception surveys, and document analysis related to QRIS security policies and practices. The findings reveal that the primary risks faced by users and business owners include QR code manipulation, social engineering attacks, unstable internet connections, and low digital literacy. Based on the identified risks, the study suggests several mitigation strategies, including the use of dynamic QRIS, user security education, infrastructure improvements, and the implementation of regular audits. In conclusion, to enhance security and user trust in QRIS, a comprehensive approach is needed, incorporating technical, procedural, and educational aspects in an integrated manner.This study aims to analyze the security risks associated with the implementation of the QRIS (Quick Response Indonesia Standard) payment system in public spaces and provide appropriate mitigation recommendations. The research employs a case study approach with a qualitative research design to explore the perceptions of users and business owners regarding the potential risks involved. Data were collected through semi-structured interviews, risk perception surveys, and document analysis related to QRIS security policies and practices. The findings reveal that the primary risks faced by users and business owners include QR code manipulation, social engineering attacks, unstable internet connections, and low digital literacy. Based on the identified risks, the study suggests several mitigation strategies, including the use of dynamic QRIS, user security education, infrastructure improvements, and the implementation of regular audits. In conclusion, to enhance security and user trust in QRIS, a comprehensive approach is needed, incorporating technical, procedural, and educational aspects in an integrated manner
Web-Based F&B Lazatto Product Sales and Stock Prediction System with Double Moving Average (DMA) Method
Get PDFThis study aims to develop a web-based sales and stock prediction system for Lazatto, a Food and Beverage (F&B) company, using the Double Moving Average (DMA) method. The background of this research is based on issues stock requirement planning is still done conventionally, where the head of the restaurant places stock orders solely based on personal experience and intuition, without utilizing past sales data as a basis for decision-making, which often result in overstocking or stockouts. By implementing a web-based forecasting information system, the company can obtain real-time and structured data. This study uses sales data from April 2024 to March 2025. The prediction results show a downward trend in sales for the "Kentang" (Potato) product, with a forecasted value of 107.33 for April 2025, compared to an actual value of 95. Model evaluation indicates an average MAPE of 21.19%, which is considered a "fair" level of forecasting accuracy. Additionally, the time required for weekly stock planning was reduced, and interviews with staff revealed increased user satisfaction and ease of use. The developed system has proven to support more accurate and efficient decision-making in inventory management
Design of an Internet of Things (IoT)-Based Fish Feeder System Using an Android Application
Get PDFFish farming plays a crucial role in aquaculture, where feed management is a key factor affecting productivity and operational costs. This research presents the design and implementation of an Internet of Things (IoT)-based automatic fish feeder system, integrated with a custom Android application. The system uses an ESP32 microcontroller to control a load cell sensor for accurate feed weighing, an ultrasonic sensor to monitor feed availability, servo motors for feed release mechanisms, and a DC motor for feed dispersion. Firebase Realtime Database serves as the data communication medium between the hardware and mobile application, enabling real-time control and monitoring. A rule-based control logic is implemented to execute scheduled or manual feeding processes. Experimental results show a feed weight accuracy of ±5 grams, with feeding operations completed within 1.5 minutes and an average throw distance of 287.8 cm. The system supports automatic alerts, scheduling, feed history logging, and remote access via the application. Compared to conventional manual methods, the system reduces feed waste, increases portion accuracy, and decreases feeding time by over 75%. These features demonstrate the system’s capability to enhance feeding efficiency, reduce labor dependency, and support sustainable and scalable fish farming practices through automation and real-time monitoring
Comparative Analysis of LightGBM and Random Forest for Daily Bitcoin Closing Price Prediction with Ensemble Approach
Get PDFThis study performs a comparative analysis of the LightGBM and Random Forest algorithms in predicting daily Bitcoin closing prices, with an exploration of an Ensemble approach for potential improvements in accuracy. A quantitative research design is employed, utilizing historical Bitcoin (BTC-USD) data from September 2015 to July 2025, enriched with various technical indicators. Data preprocessing, model training, and evaluation were carried out using Python in Google Colaboratory, with the dataset split into 80% for training and 20% for testing. Model performance was evaluated using the Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the R-squared (R²) statistic, with statistical significance tests to ensure robust comparisons. A simple Linear Regression model was also included as a baseline. The findings reveal that Random Forest outperformed LightGBM, achieving an MAE of 11,599.74, an RMSE of 19,262.31, and an R² of 0.431, compared to LightGBM’s MAE of 12,285.42, RMSE of 19,995.04, and R² of 0.386. Although the Ensemble model showed slight improvements over LightGBM, it did not surpass Random Forest. The relatively low R² values across all models reflect the inherent volatility and difficulty in predicting Bitcoin prices. The study concludes that Random Forest demonstrates superior robustness for Bitcoin forecasting. Importantly, this work provides a novel empirical contribution by being one of the first to directly benchmark RF, LightGBM, and their Ensemble for Bitcoin prediction, highlighting that a simple averaging Ensemble does not guarantee superior performance. This finding provides a foundation for developing more refined Ensemble strategies tailored to high-volatility assets
Gaussian Mixture-Based Data Augmentation Improves QSAR Prediction of Corrosion Inhibition Efficiency
Get PDFPredicting corrosion inhibition efficiency IE (%) is often hindered by small, heterogeneous datasets. This study proposes a Gaussian mixture–based data augmentation pipeline to strengthen QSAR generalization under data scarcity. A curated set of 70 drug-like compounds with 14 physicochemical and quantum descriptors was cleaned, split 90/10 (train/test), and transformed using a Quantile Transformer followed by a Robust Scaler. A Gaussian Mixture model (GMM) with 2–5 components selected by the variational lower bound was fitted to the transformed training features and used to generate up to 2,500 synthetic samples. Eight regressors (Gaussian Process, Decision Tree, Random Forest, Bagging, Gradient Boosting, Extra Trees, SVR, and KNN) were evaluated on the held-out test set using R2 and RMSE. Augmentation improved performance across several families: for example, Gaussian Process R2 improved from −1.54 to 0.54 (RMSE 11.71 to 5.01) and Decision Tree R2 from −0.33 to 0.63 (RMSE 8.48 to 4.44), Bagging and Random Forest showed R2 increases of 0.67 and 0.40, respectively. The optimal synthetic size varied by model
Face Recognition Using MTCNN Face Detection, ResNetV1 Feature Embeddings, and SVM Classification
Get PDFFace recognition has become an essential component of modern security and authentication systems, yet its effectiveness is often challenged by limited datasets, class imbalance, variations in facial poses, lighting conditions, and image resolutions. This study proposes a face recognition pipeline that integrates Multi-task Cascaded Convolutional Networks (MTCNN) for face detection, Residual Network V1 (ResNetV1) for feature extraction, and Support Vector Machine (SVM) for classification. Unlike previous works that rely on large-scale datasets and end-to-end deep learning models, this study emphasizes the effectiveness of the pipeline under constrained data conditions, using 856 images across 191 classes with highly imbalanced distribution. Experimental results show that MTCNN successfully detected 97.1% of faces, while ResNetV1 produced 512-dimensional embeddings that formed well-separated clusters validated by clustering metrics (Silhouette Score = 0.578, Davies-Bouldin Index = 0.566). The SVM classifier achieved 92.9% accuracy, with macro-average precision, recall, and F1-scores of 0.89, 0.92, and 0.89 respectively, significantly outperforming a baseline k-Nearest Neighbor (k-NN) model that only reached 63.9% accuracy. These findings highlight the novelty of this study: demonstrating that a lightweight yet robust pipeline can deliver reliable recognition performance even in small, imbalanced datasets, making it suitable for real-world scenarios where large-scale training data are not available