Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Stacking of DT, RF, and Gradient Boosting Algorithms for Classification of Building Damage Due to Earthquakes
No full textClassification of building damage levels due to earthquakes is an important aspect in disaster mitigation and post-disaster risk assessment. This study aims to improve classification accuracy on imbalanced data using an ensemble stacking method. It combines Decision Tree, Random Forest, and Gradient Boosting algorithms, with Logistic Regression as a meta-learner. The building damage dataset from the 2015 Gorkha Nepal earthquake underwent data cleaning, categorical transformation, normalization, and balancing using ADASYN. Evaluation showed that Random Forest was the best single model. The stacking model achieved the highest accuracy of 91.77% after balancing. These results show that stacking improves generalization and classification accuracy on imbalanced data. This suggests significant potential for integration into disaster decision-support systems that require fast, accurate building-damage assessment
Evaluation of SMOTE Technique in the Comparison of XGBoost and Random Forest Algorithms for Liver Disease Prediction
No full textIn many countries, including Indonesia, liver disease remains a major cause of morbidity and mortality. Early detection plays a crucial role in improving treatment outcomes. This study evaluates the performance of two widely used machine learning models Random Forest and XGBoost for predicting liver disease, employing the SMOTE balancing technique to address class imbalance. The primary objectives are to enhance model fairness, reduce overfitting, and improve sensitivity toward the minority class. Model performance is assessed using accuracy, precision, recall, and F1-score metrics. The XGBoost model achieved an average accuracy of 99.74%, precision of 99.77%, recall of 99.75%, and F1-score of 99.72%, while the Random Forest model attained an average accuracy of 99.82%, precision of 99.89%, recall of 99.75%, and F1-score of 99.75%. Both models demonstrated excellent predictive capability, with Random Forest slightly outperforming XGBoost. These results highlight the importance of data balancing and robust model validation in developing reliable machine learning models for healthcare decision-making.In many countries, including Indonesia, liver disease remains a major cause of morbidity and mortality. Early detection plays a crucial role in improving treatment outcomes. This study evaluates the performance of two widely used machine learning models Random Forest and XGBoost for predicting liver disease, employing the SMOTE balancing technique to address class imbalance. The primary objectives are to enhance model fairness, reduce overfitting, and improve sensitivity toward the minority class. Model performance is assessed using accuracy, precision, recall, and F1-score metrics. The XGBoost model achieved an average accuracy of 99.74%, precision of 99.77%, recall of 99.75%, and F1-score of 99.72%, while the Random Forest model attained an average accuracy of 99.82%, precision of 99.89%, recall of 99.75%, and F1-score of 99.75%. Both models demonstrated excellent predictive capability, with Random Forest slightly outperforming XGBoost. These results highlight the importance of data balancing and robust model validation in developing reliable machine learning models for healthcare decision-making
Stock Price Prediction Using Deep Learning (LSTM) with a Recursive Approach
No full textStock price prediction is a critical topic in financial technology research, as accurate forecasts support better decision-making in volatile markets. Numerous studies have applied statistical and machine learning models; however, most focus on one-step-ahead predictions and lack evaluation of recursive strategies in multi-day horizons. This study investigates the application of Long Short-Term Memory (LSTM) with a recursive forecasting approach to enhance stock price prediction accuracy. The dataset was enriched with multiple technical indicators and processed through a systematic Knowledge Discovery in Databases (KDD) pipeline, including preprocessing, transformation, modelling, and evaluation. Experimental results show that the recursive LSTM model achieves superior performance compared to baseline machine learning methods, with high accuracy in short-term horizons and stable performance up to nine days ahead, although accuracy gradually declines due to error accumulation. This work highlights the importance of integrating recursive forecasting with technical indicators to improve predictive capability in emerging markets and provides a foundation for developing adaptive financial forecasting frameworks
Image-Based Classification of Indonesian Traditional Houses Using a Hybrid CNN-SVM Algorithm
No full textThe diversity of Indonesian traditional houses represents a cultural heritage that must be preserved. However, the lack of interest among younger generations and the difficulty in recognizing the distinctive architectural characteristics of traditional houses present challenges to preservation efforts. This study aims to develop an image classification model for Indonesian traditional houses using a hybrid CNN-SVM approach to improve recognition accuracy. The dataset consists of 3,919 images from five classes of traditional houses, namely gadang, joglo, panjang, tongkonan, and honai, with an 80% training split, 10% validation, and 10% testing. The data were processed through resizing, augmentation, and normalization before being trained using a CNN architecture with five convolutional layers as a feature extractor and an SVM serving as a multi-class classifier. The experimental results show that the hybrid CNN-SVM model achieved an accuracy of 96.68%, with consistently high precision, recall, and F1-score across all classes. These findings demonstrate that integrating CNN as a feature extractor and SVM as the final classifier can enhance the model’s generalization capability in distinguishing images of Indonesian traditional houses.The diversity of Indonesian traditional houses represents a cultural heritage that must be preserved. However, the lack of interest among younger generations and the difficulty in recognizing the distinctive architectural characteristics of traditional houses present challenges to preservation efforts. This study aims to develop an image classification model for Indonesian traditional houses using a hybrid CNN-SVM approach to improve recognition accuracy. The dataset consists of 3,919 images from five classes of traditional houses, namely gadang, joglo, panjang, tongkonan, and honai, with an 80% training split, 10% validation, and 10% testing. The data were processed through resizing, augmentation, and normalization before being trained using a CNN architecture with five convolutional layers as a feature extractor and an SVM serving as a multi-class classifier. The experimental results show that the hybrid CNN-SVM model achieved an accuracy of 96.68%, with consistently high precision, recall, and F1-score across all classes. These findings demonstrate that integrating CNN as a feature extractor and SVM as the final classifier can enhance the model’s generalization capability in distinguishing images of Indonesian traditional houses
Ethical Analysis of Online Media Journalistic Photos Worth Publishing Based on Images Using the Convolutional Neural Network Method
No full textThis study aims to develop and test a Convolutional Neural Network (CNN)-based artificial intelligence model to analyze and classify online media journalistic photos based on ethical criteria for publication suitability (suitable or unsuitable). In the context of digital journalism, the process of filtering sensitive visual content that potentially violates the code of ethics is often time-consuming and prone to subjectivity. Therefore, a CNN model is proposed as an automated solution to identify images containing visual elements deemed unethical. An annotated image dataset was used to train and test the CNN model. The model test results showed effective and robust performance in classifying the ethical suitability of photos. The model achieved a weighted average accuracy of 0.86 (86%) and a weighted average F1 - score of 0.86. Specifically, the model performed very well in identifying "suitable" photos with precision, recall, and F1- score values ranging from 0.88 to 0.89. Performance in the "Unsuitable" class was also relatively strong with an F1 - score of 0.81. Overall, these results confirm that the CNN method has great potential as an efficient and objective decision support system in the visual content editing process. Implementing this model not only speeds up the editorial process but also improves online media\u27s adherence to journalistic ethical standards by minimizing the risk of publishing potentially unethical photos.This study aims to develop and test a Convolutional Neural Network (CNN)-based artificial intelligence model to analyze and classify online media journalistic photos based on ethical criteria for publication suitability (suitable or unsuitable). In the context of digital journalism, the process of filtering sensitive visual content that potentially violates the code of ethics is often time-consuming and prone to subjectivity. Therefore, a CNN model is proposed as an automated solution to identify images containing visual elements deemed unethical. An annotated image dataset was used to train and test the CNN model. The model test results showed effective and robust performance in classifying the ethical suitability of photos. The model achieved a weighted average accuracy of 0.86 (86%) and a weighted average F1 - score of 0.86. Specifically, the model performed very well in identifying "suitable" photos with precision, recall, and F1- score values ranging from 0.88 to 0.89. Performance in the "Unsuitable" class was also relatively strong with an F1 - score of 0.81. Overall, these results confirm that the CNN method has great potential as an efficient and objective decision support system in the visual content editing process. Implementing this model not only speeds up the editorial process but also improves online media\u27s adherence to journalistic ethical standards by minimizing the risk of publishing potentially unethical photos
Performance Analysis of the Item-Based Collaborative Filtering Model in Yogyakarta Tourism Recommendations
Get PDFYogyakarta is one of the most popular tourist destinations in Indonesia, offering a variety of attractions ranging from beaches and mountains to historical sites. This diversity poses a challenge for tourists in selecting destinations that match their preferences. This study employs the Item-Based Collaborative Filtering method to recommend tourist destinations based on the similarity between attractions, calculated using cosine similarity. The data analyzed includes 1,069 tourist destinations in Yogyakarta, obtained from Google Maps API, Scrapetable, and Outscraper. The results indicate that the developed recommendation model achieves high accuracy with a Mean Absolute Error (MAE) of 2.537. Compared to previous approaches, this method improves the relevance and quality of recommendations, helping tourists find destinations that suit their preferences. This study contributes to the development of more personalized and effective recommendation systems for the tourism sector.Yogyakarta is one of the most popular tourist destinations in Indonesia, offering a variety of attractions ranging from beaches and mountains to historical sites. This diversity poses a challenge for tourists in selecting destinations that match their preferences. This study employs the Item-Based Collaborative Filtering method to recommend tourist destinations based on the similarity between attractions, calculated using cosine similarity. The data analyzed includes 1,069 tourist destinations in Yogyakarta, obtained from Google Maps API, Scrapetable, and Outscraper. The results indicate that the developed recommendation model achieves high accuracy with a Mean Absolute Error (MAE) of 2.537. Compared to previous approaches, this method improves the relevance and quality of recommendations, helping tourists find destinations that suit their preferences. This study contributes to the development of more personalized and effective recommendation systems for the tourism sector
Evaluation of Wireless LAN Quality of Service (QoS) in Primary Education Using TIPHON Standards
Get PDFIn the current digital era, internet connectivity in schools is crucial to support teaching and learning activities. SD Negeri 2 Sumber has implemented a wireless LAN network to provide internet access for students and teachers. This study aims to evaluate the network performance by measuring Quality of Service or QoS parameters, namely throughput, packet loss, delay, and jitter. The evaluation was conducted using Wireshark to monitor network traffic. The results show that the average throughput for video streaming is 4.251 Kbps, browsing is 1.425 Kbps, and downloading is 3.106 Kbps. The average packet loss is 1.66 % for video streaming, 4.6 % for browsing, and 2.66 % for downloading. The average delay for video streaming is 1.64 ms, browsing is 5.92 ms, and downloading is 2.32 ms. The average jitter is 1.62 ms for video streaming, 5.92 ms for browsing, and 2.16 ms for downloading. Based on the QoS parameters, the network quality is categorized as good according to TIPHON standards with a final score of 3.75. Although the overall network quality is good, there are several areas that need optimization, such as browsing activities, which show slightly higher throughput and jitter compared to other activities. This study provides a clear overview of wireless network performance and offers recommendations for further optimization to enhance user experience, particularly in activities involving browsing and school administration.In the current digital era, internet connectivity in schools is crucial to support teaching and learning activities. SD Negeri 2 Sumber has implemented a wireless LAN network to provide internet access for students and teachers. This study aims to evaluate the network performance by measuring Quality of Service or QoS parameters, namely throughput, packet loss, delay, and jitter. The evaluation was conducted using Wireshark to monitor network traffic. The results show that the average throughput for video streaming is 4.251 Kbps, browsing is 1.425 Kbps, and downloading is 3.106 Kbps. The average packet loss is 1.66 % for video streaming, 4.6 % for browsing, and 2.66 % for downloading. The average delay for video streaming is 1.64 ms, browsing is 5.92 ms, and downloading is 2.32 ms. The average jitter is 1.62 ms for video streaming, 5.92 ms for browsing, and 2.16 ms for downloading. Based on the QoS parameters, the network quality is categorized as good according to TIPHON standards with a final score of 3.75. Although the overall network quality is good, there are several areas that need optimization, such as browsing activities, which show slightly higher throughput and jitter compared to other activities. This study provides a clear overview of wireless network performance and offers recommendations for further optimization to enhance user experience, particularly in activities involving browsing and school administration
The Role of Marketplaces in Assisting MSME Growth Using the Systematic Literature Review Method
Get PDFDigital technology provides significant opportunities for Micro, Small, and Medium Enterprises (MSMEs) to grow and innovate, contributing to national economic growth. This study aims to analyze the role, impact, and supporting factors of marketplaces in fostering MSME development in Indonesia. Using the Systematic Literature Review (SLR) method, 103 journals were identified through Publish or Perish via Crossref for the 2019–2024 period. These journals were then filtered down to 40 journals selected for descriptive analysis. The 2019–2024 timeframe was chosen as it reflects a significant surge in marketplace adoption during the COVID-19 pandemic.The findings reveal that marketplaces significantly enhance MSMEs by increasing market reach by up to 40%, improving operational efficiency through a 30% reduction in order processing time, and building customer trust through reviews and ratings. The study also highlights the importance of supporting factors such as adequate digital infrastructure, digital literacy, and support from the government and platform providers. However, challenges such as limited internet access, low digital literacy, high commission fees, and intense competition with larger brands hinder the optimal use of marketplaces by MSMEs.This study recommends several practical solutions, including strengthening digital infrastructure, subsidizing internet costs, providing digital literacy training, and offering special promotion programs for small MSMEs. Although this study relies on secondary data, the findings provide relevant insights for policymakers, platform providers, and MSME actors. Further studies using primary data are needed to enhance the generalizability of these findings.Digital technology provides significant opportunities for Micro, Small, and Medium Enterprises (MSMEs) to grow and innovate, contributing to national economic growth. This study aims to analyze the role, impact, and supporting factors of marketplaces in fostering MSME development in Indonesia. Using the Systematic Literature Review (SLR) method, 103 journals were identified through Publish or Perish via Crossref for the 2019–2024 period. These journals were then filtered down to 40 journals selected for descriptive analysis. The 2019–2024 timeframe was chosen as it reflects a significant surge in marketplace adoption during the COVID-19 pandemic.The findings reveal that marketplaces significantly enhance MSMEs by increasing market reach by up to 40%, improving operational efficiency through a 30% reduction in order processing time, and building customer trust through reviews and ratings. The study also highlights the importance of supporting factors such as adequate digital infrastructure, digital literacy, and support from the government and platform providers. However, challenges such as limited internet access, low digital literacy, high commission fees, and intense competition with larger brands hinder the optimal use of marketplaces by MSMEs.This study recommends several practical solutions, including strengthening digital infrastructure, subsidizing internet costs, providing digital literacy training, and offering special promotion programs for small MSMEs. Although this study relies on secondary data, the findings provide relevant insights for policymakers, platform providers, and MSME actors. Further studies using primary data are needed to enhance the generalizability of these findings
Expert System for Diagnosing Shallot Plant Diseases Using the Forward Chaining Method
Get PDFThis research aims to develop an expert system with the forward chaining method to diagnose diseases in shallot plants in Mijen District, Demak Regency. The decline in shallot production caused by disease is one of the main problems faced by local farmers. Therefore, this expert system was developed to help farmers diagnose diseases in their crops more quickly and accurately.The research method used combines qualitative and quantitative approaches. Data on plant diseases and symptoms were obtained through interviews with agricultural extension workers and literature review. Based on the data, a knowledge base was built that was used in the expert system. Forward chaining was applied to trace the relationship between symptoms inputted by the user and possible disease diagnoses. The system was tested using validation data, with an accuracy result of 93.3%, indicating that the system has a high level of agreement with the diagnosis provided by the expert.The results of this study show that the developed web-based expert system can provide practical solutions for farmers to diagnose and treat diseases in shallot plants, so as to increase agricultural productivity and reduce losses due to disease attacks. With web-based implementation, this system can be easily accessed by farmers through computer or mobile devices, providing ease of use in the field
Magnetic Resonance Imaging for Breast Cancer Classification Using Convolutional Neural Networks
Get PDFBreast cancer remains a leading cause of mortality among women worldwide, emphasizing the urgent need for accurate diagnostic methods. This research addresses the challenges of early detection by leveraging Convolutional Neural Networks (CNNs) for the classification of Magnetic Resonance Imaging (MRI) data. Using a publicly available Kaggle dataset consisting of 54,676 MRI images categorized into "Normal" and "Cancer" classes, the dataset was split into 80% for training and 20% for validation. A modified CNN architecture was developed, incorporating optimized layers and hyperparameters, such as the ADAM optimizer, a learning rate of 0.0001, and a mini-batch size of 128. The proposed model achieved exceptional performance, with an accuracy of 99.72%, precision and recall of 99.98% and 99.97%, respectively, and an F1-score of 99.98%, as evaluated through a confusion matrix. These results demonstrate the model’s robustness in distinguishing between healthy and cancerous tissues, providing a reliable and efficient diagnostic tool. This study highlights the potential of CNNs to improve diagnostic precision in medical imaging, aiding clinicians and advancing AI applications in healthcare