Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Comparison of Naive Bayes and Support Vector Machine (SVM) Algorithms for Classifying the Maturity Level of Melon
No full textThis determination of melon fruit ripeness is an important factor in ensuring fruit quality in terms of taste, texture, and market value. However, ripeness assessment is still predominantly performed manually and relies on subjective judgement, which may lead to decreased product quality, inefficient distribution processes, and potential economic losses. Therefore, an automated approach for classifying melon ripeness levels is required. This study aims to analyze and compare the performance Support Vector Machine (SVM) and Naïve Bayes algorithms for melon ripeness classification based on digital images using Histogram of Oriented Gradients (HOG) feature extraction method. The dataset used in this study consists of 630 melon images divided into three ripeness classes, 209 unripe, 220 semi ripe, and 201 ripe images. The research process includes image preprocessing, data augmentation, feature extraction, model training, and performance evaluation. Experimental results show that the SVM with a Radial Basis Function (RBF) kernel, using parameter C=10 and the default value, achieves the highest classification accuracy of 94%, while the Naïve Bayes algorithm attains an accuracy of 65%. These results indicate that the SVM algorithm demonstrates superior classification performance compared to Naïve Bayes in determining melon ripeness levels.This determination of melon fruit ripeness is an important factor in ensuring fruit quality in terms of taste, texture, and market value. However, ripeness assessment is still predominantly performed manually and relies on subjective judgement, which may lead to decreased product quality, inefficient distribution processes, and potential economic losses. Therefore, an automated approach for classifying melon ripeness levels is required. This study aims to analyze and compare the performance Support Vector Machine (SVM) and Naïve Bayes algorithms for melon ripeness classification based on digital images using Histogram of Oriented Gradients (HOG) feature extraction method. The dataset used in this study consists of 630 melon images divided into three ripeness classes, 209 unripe, 220 semi ripe, and 201 ripe images. The research process includes image preprocessing, data augmentation, feature extraction, model training, and performance evaluation. Experimental results show that the SVM with a Radial Basis Function (RBF) kernel, using parameter C=10 and the default value, achieves the highest classification accuracy of 94%, while the Naïve Bayes algorithm attains an accuracy of 65%. These results indicate that the SVM algorithm demonstrates superior classification performance compared to Naïve Bayes in determining melon ripeness levels
Knowledge Discovery in Sharia Mobile Banking Reviews Using Aspect-Based Sentiment Analysis and Machine Learning
No full textUser reviews provide important insights into the quality of digital banking applications; however, their large volume makes manual analysis inefficient. This study applies Aspect-Based Sentiment Analysis (ABSA) to examine user perceptions of the BYOND by BSI application based on three aspects: interface, features and performance, and services. Three classification algorithms were compared: Naïve Bayes, Support Vector Machine (SVM), and Random Forest, evaluated with accuracy, precision, recall, F1-score, and ROC-AUC. The results indicate that SVM and Naïve Bayes achieved the best performance, with an accuracy of 0.95 and an F1-score of 0.92, whereas Random Forest exhibited slightly lower performance with an F1-score of 0.89. Furthermore, sentiment analysis reveals the features and performance aspect exhibits the highest proportion of negative sentiment (39.6%), primarily associated with system reliability issues, login problems, transaction failures, and application instability. These findings demonstrate that ABSA can serve as an effective knowledge discovery approach for identifying critical functional issues and supporting data-driven prioritization in improving digital banking services, particularly within the context of sharia banking applications.User reviews provide important insights into the quality of digital banking applications; however, their large volume makes manual analysis inefficient. This study applies Aspect-Based Sentiment Analysis (ABSA) to examine user perceptions of the BYOND by BSI application based on three aspects: interface, features and performance, and services. Three classification algorithms were compared: Naïve Bayes, Support Vector Machine (SVM), and Random Forest, evaluated with accuracy, precision, recall, F1-score, and ROC-AUC. The results indicate that SVM and Naïve Bayes achieved the best performance, with an accuracy of 0.95 and an F1-score of 0.92, whereas Random Forest exhibited slightly lower performance with an F1-score of 0.89. Furthermore, sentiment analysis reveals the features and performance aspect exhibits the highest proportion of negative sentiment (39.6%), primarily associated with system reliability issues, login problems, transaction failures, and application instability. These findings demonstrate that ABSA can serve as an effective knowledge discovery approach for identifying critical functional issues and supporting data-driven prioritization in improving digital banking services, particularly within the context of sharia banking applications
Exploring Public Opinion on the \u27Makan Bergizi Gratis\u27 Program on X: A Comparative Analysis of IndoBERT-Large and NusaBERT-Large Models
No full textProgram Makan Bergizi Gratis (MBG) has triggered extensive discourse on social media platform X, which serves as a primary space for public expression of opinions toward government policies. This study aims to analyze public sentiment toward the MBG program while simultaneously comparing the performance of two prominent Transformer-based models, namely IndoBERT-Large and NusaBERT-Large. This research adopts a quantitative approach employing supervised learning on 10,201 Indonesian-language posts (tweets) collected through web scraping from February 2024 to September 2025. A total of 2,000 samples were manually annotated as ground truth, achieving a high level of inter-annotator reliability (Cohen’s Kappa, κ = 0.81). The experimental results indicate that IndoBERT-Large outperforms NusaBERT-Large, achieving an accuracy of 83.00%, while NusaBERT-Large demonstrates competitive performance with an accuracy of 80.50%. Substantively, public discourse is dominated by negative sentiment, accounting for nearly 50% of the total data, reflecting public concerns regarding budgetary constraints and technical implementation issues. Positive sentiment ranges between 33% and 36%, indicating sustained and substantial public support for the program. These findings confirm the effectiveness of Transformer-based models in accurately capturing the dynamics of public opinion toward government policies using social media data.Program Makan Bergizi Gratis (MBG) has triggered extensive discourse on social media platform X, which serves as a primary space for public expression of opinions toward government policies. This study aims to analyze public sentiment toward the MBG program while simultaneously comparing the performance of two prominent Transformer-based models, namely IndoBERT-Large and NusaBERT-Large. This research adopts a quantitative approach employing supervised learning on 10,201 Indonesian-language posts (tweets) collected through web scraping from February 2024 to September 2025. A total of 2,000 samples were manually annotated as ground truth, achieving a high level of inter-annotator reliability (Cohen’s Kappa, κ = 0.81). The experimental results indicate that IndoBERT-Large outperforms NusaBERT-Large, achieving an accuracy of 83.00%, while NusaBERT-Large demonstrates competitive performance with an accuracy of 80.50%. Substantively, public discourse is dominated by negative sentiment, accounting for nearly 50% of the total data, reflecting public concerns regarding budgetary constraints and technical implementation issues. Positive sentiment ranges between 33% and 36%, indicating sustained and substantial public support for the program. These findings confirm the effectiveness of Transformer-based models in accurately capturing the dynamics of public opinion toward government policies using social media data
Performance Evaluation of Web Applications Using JMeter Load Testing for Server Capacity and Response Efficiency
No full textThe reliability of web-based conference systems is crucial for ensuring smooth services during periods of high activity. This research evaluates the performance of the ICOMTA PPNS journal website by conducting load testing using Apache JMeter with scenarios ranging from 50 to 5000 virtual users, each executed for one hour. The evaluation focuses on response time, error rate, throughput, and bandwidth usage. The results indicate that the website performs reliably with up to 200 concurrent users, demonstrating stable response times and no recorded errors. However, once the load surpasses 300 users, response times increase sharply exceeding 60 seconds and errors begin to appear, suggesting that the server has reached its performance limit. Under the heaviest load of 5000 users, throughput continues to rise, but overall service quality declines significantly. These findings highlight the need for server enhancements or migration to cloud-based infrastructure to ensure stable performance during peak usage
Anomaly-Based DDoS Detection Using Improved Deep Support Vector Data Description (Deep SVDD) and Multi-Model Ensemble Approach
No full textDistributed Denial-of-Service (DDoS) attacks remain a critical threat to network infrastructure, demanding robust and efficient detection mechanisms. This study proposes an enhanced Deep Support Vector Data Description (Deep SVDD) model for unsupervised DDoS detection using the UNSW-NB15 dataset. The approach leverages a deep encoder architecture with batch normalization and dropout to learn compact latent representations of normal traffic, minimizing the hypersphere volume enclosing benign flows. Only normal samples are used during training, adhering to the unsupervised anomaly detection paradigm. The model is evaluated against five established baselines—Isolation Forest, Local Outlier Factor (LOF), One-Class SVM, Autoencoder, and a simple ensemble—using AUC, F1-score, and recall as primary metrics. Experimental results demonstrate that Deep SVDD significantly outperforms all baselines, achieving superior class separation, high detection sensitivity, and computational efficiency (0.0004 GFLOPs). Notably, while LOF exhibited a deceptively high F1-score, its AUC near 0.5 revealed poor discriminative capability, highlighting the risk of relying on single metrics. The ensemble approach failed to improve performance, underscoring the limitation of naive score averaging when weak detectors are included. Visualization of score distributions and ROC curves further confirms Deep SVDD’s ability to effectively distinguish DDoS from benign traffic. These findings affirm that representation learning in latent space offers a more reliable foundation for anomaly detection than traditional distance-, density-, or reconstruction-based methods. The proposed model presents a promising solution for real-time, low-overhead intrusion detection systems in modern network environments. Future work will explore adaptive ensembles, self-supervised pretraining, and deployment on edge devices
Analysis of BRIsat Investment Success from Financial and Nonfinancial Perspectives
No full textInvestment in information technology within the banking sector requires not only financial viability but also public acceptance, particularly for state-owned enterprises (SOEs) that carry both commercial and social mandates. This study aims to evaluate the success of PT Bank Rakyat Indonesia (Persero) Tbk.’s BRIsat satellite investment from both financial and nonfinancial perspectives. A Cost–Benefit Analysis (CBA) was employed to assess financial feasibility using BRI’s publicly available financial statements from 2014 to 2024, while sentiment analysis using the Naive Bayes algorithm was conducted to examine public perception based on social media data from platform X covering the period 2016–2024. The financial analysis indicates that the BRIsat investment is financially feasible, with a Return on Investment (ROI) of 2.58%, a Payback Period of 6.2 years, a positive Net Present Value (NPV) of IDR 166,161,960, and a Benefit Cost Ratio (BCR) of 184.9, suggesting that every IDR 1 invested generates IDR 184.9 in economic benefits. From the nonfinancial perspective, sentiment analysis of 10,066 valid tweets reveals that 55.90% of public sentiment is negative (5,627 tweets), while 44.10% is positive (4,439 tweets), with the Naive Bayes model achieving an accuracy of 96.76%. Positive sentiment is primarily associated with keywords such as “successful,” “fast,” and “service,” reflecting appreciation for BRIsat as a strategic innovation, whereas negative sentiment is dominated by terms such as “error,” “failed,” and “disruption,” indicating persistent technical issues in digital banking services. These findings highlight a clear contradiction between the strong financial performance of the BRIsat investment and the predominantly negative public perception of service quality. The study implies that the success of large-scale technology investments in SOEs cannot be assessed solely through financial metrics, but must be accompanied by continuous improvements in operational reliability and digital service quality to ensure sustainable value creation and public trust
English English
No full textThis research aims to compare the performance of the Apriori and FP-Growth algorithms in the process of data mining association patterns in the online sales transaction data of a bookstore. The dataset used consists of 74.090 transactions resulting from data cleaning from the period January-June 2025. The analysis was conducted through the stages of data collection, followed by data preparation consisting of data cleaning and data transformation, and then continued to the modeling stage of the two algorithms. The results of the experiment show that Apriori tends to be faster on small-scale datasets with simple transaction patterns, while FP-Growth has more stable memory usage and shows more efficient processing time on parameters that analyze larger data. Both algorithms produce identical numbers and contents of association rules for each parameter variation, indicating that the significant difference lies in performance efficiency, and not in the knowledge patterns produced. Rules with the highest lift values, such as the association between the books "Rumah Kaca" and "Jejak Langkah" (lift: 183,306 & confidence 0,903) and between the books "Namaku Alam" and "Pulang" (lift: 34,062 & confidence: 0,51) indicate strong purchasing patterns between titles with the same author and theme. These findings have the potential to support cross-selling strategies and product recommendations in online sales systems. This research is still limited to a relatively small and homogeneous dataset, so further using a broader data coverage is recommended to test the algorithm\u27s performance more comprehensively
Suicidal Ideation Detection in Social Media using Optimized CNN-BiLSTM Architecture
No full textThis research aims to develop an optimized hybrid deep learning model for detecting suicidal ideation from social media text. The growing volume of online discussions, particularly on platforms such as Reddit, provides valuable signals for early identification of individuals at risk; however, the linguistic characteristics of user-generated content are highly diverse and often noisy. To address this challenge, this study proposes an Optimized CNN-BiLSTM architecture enhanced with a dropout rate of 0.6 and a strategic training approach utilizing Early Stopping (patience=3) and a Learning Rate Scheduler (ReduceLROnPlateau) to prevent local minima and ensure convergence stability. The dataset used consists of 232,074 text entries with a balanced class distribution (50% suicide, 50% non-suicide) to ensure the validity of evaluation metrics and eliminate majority class bias. Experimental results demonstrate that the optimized model achieves an accuracy of 94.96%, precision of 95.70%, recall of 94.15%, and an F1-score of 94.92%, indicating a significant improvement over the baseline CNN-BiLSTM and single BiLSTM models. Furthermore, interpretability analysis via keyword visualization (Word Cloud) validates that the model effectively captures semantically relevant emotional expressions of despair. These findings suggest that the optimized hybrid architecture provides a robust and operationally viable approach for supporting real-time early-warning systems on social media platforms to facilitate timely mental health interventions.This research aims to develop an optimized hybrid deep learning model for detecting suicidal ideation from social media text. The growing volume of online discussions, particularly on platforms such as Reddit, provides valuable signals for early identification of individuals at risk; however, the linguistic characteristics of user-generated content are highly diverse and often noisy. To address this challenge, this study proposes an Optimized CNN-BiLSTM architecture enhanced with a dropout rate of 0.6 and a strategic training approach utilizing Early Stopping (patience=3) and a Learning Rate Scheduler (ReduceLROnPlateau) to prevent local minima and ensure convergence stability. The dataset used consists of 232,074 text entries with a balanced class distribution (50% suicide, 50% non-suicide) to ensure the validity of evaluation metrics and eliminate majority class bias. Experimental results demonstrate that the optimized model achieves an accuracy of 94.96%, precision of 95.70%, recall of 94.15%, and an F1-score of 94.92%, indicating a significant improvement over the baseline CNN-BiLSTM and single BiLSTM models. Furthermore, interpretability analysis via keyword visualization (Word Cloud) validates that the model effectively captures semantically relevant emotional expressions of despair. These findings suggest that the optimized hybrid architecture provides a robust and operationally viable approach for supporting real-time early-warning systems on social media platforms to facilitate timely mental health interventions
Identification and Classification of Cracks in Traditional Pottery from West Sumatra Using Digital Image Processing
No full textCracks in traditional West Sumatran pottery are a major challenge in preserving this cultural heritage. With age and the manual manufacturing process, pottery becomes highly susceptible to physical damage, particularly cracks on the surface and internal structure. These cracks not only affect the functional and aesthetic value but also reduce the cultural and economic value of the pottery. Therefore, an accurate early identification system is crucial to ensure the survival and preservation of this culture. This study developed a digital image processing-based system to detect and classify cracks in traditional pottery. The system integrates image preprocessing, including cropping, resizing, grayscale conversion, contrast stretching, and histogram equalization to improve image quality and highlight thin and irregular cracks. Image segmentation was performed using the Multi-Threshold Otsu method to separate cracks from the background, while classification was performed using a convolutional neural network (CNN). Experimental results show that this system is able to achieve an accuracy of 94.8%, precision of 93.5%, recall of 92.3%, and F1-score of 92.9%, indicating the system\u27s ability to accurately detect cracks. Comparisons with other segmentation and classification methods are needed to provide a more comprehensive picture of the effectiveness of this approach. The implementation of this system is expected to support the preservation of traditional Minangkabau pottery through digitalization, provide an ornament database that can be accessed by researchers, artists, and the general public, and assist in more efficient cultural documentation and archiving
Image Processing and Object Detection in the Indonesian Sign System (SIBI) for Hearing-Impaired Communication
No full textCommunication is a fundamental human need, yet individuals with hearing impairments continue to face barriers due to limited access to sign language translation technologies. In Indonesia, the adoption of such technologies remains low, particularly in regions such as Sorong, Southwest Papua, creating a communication gap between the Deaf community and the general public. This study develops a web-based detection system for 36 classes of the Indonesian Sign System (SIBI) using the YOLOv5 algorithm. The dataset consists of 5,682 images of SIBI hand poses with variations in lighting and background, divided into 4,970 training images (87%), 376 validation images (7%), and 335 test images (6%). All data were processed through labeling, preprocessing, augmentation, balancing, and model training. The training was conducted for 150 epochs, and the evaluation results show that YOLOv5 is capable of detecting SIBI signs with significant accuracy. Performance evaluation using a confusion matrix achieved a detection accuracy of 95%, supported by stable precision and recall values and real-time inference performance on common web browsers. Usability testing with 20 respondents indicated satisfaction levels above 72.8%, demonstrating that the system is practical and easy to use. This research presents a validated real-time, web-based SIBI detection system that supports inclusive computer vision applications and enhances accessibility in public services such as education, healthcare, and administrative environments