Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Machine Learning Based Prediction of Osteoporosis Risk Using the Gradient Boosting Algorithm and Lifestyle Data
No full textOsteoporosis is a degenerative disease characterized by decreased bone mass and an increased risk of fractures, particularly among the elderly population. Early detection is essential; however, standard diagnostic methods such as Dual-Energy X-ray Absorptiometry (DEXA) remain limited in terms of availability and cost. This study aims to develop a machine learning-based risk prediction model for osteoporosis by utilizing lifestyle data with the Gradient Boosting algorithm. The secondary dataset was obtained from the Kaggle platform, consisting of 1,958 samples covering lifestyle and clinical attributes such as age, gender, physical activity, smoking habits, calcium intake, vitamin D consumption, and family history. Preprocessing involved normalization and categorical feature encoding, along with a balance check of class distribution, which indicated that the dataset was relatively balanced. The data were then divided using stratified sampling with an 80% training set and 20% testing set. Model performance was evaluated using accuracy, precision, recall, F1-score, and the area under the ROC curve (AUC). The results showed that the Gradient Boosting algorithm achieved an accuracy of 91%, precision of 90.8%, recall of 90.2%, F1-score of 90.5%, and an AUC of 0.92, outperforming baseline methods such as Logistic Regression and Random Forest. These findings demonstrate that Gradient Boosting is effective as a decision-support tool for early osteoporosis screening based on lifestyle data and has the potential to be integrated into clinical decision-making systems to enhance early detection in healthcare services. Nevertheless, since this study relied on a secondary dataset from Kaggle, the results require further validation using real clinical data from Indonesia to ensure representativeness for the local population
Sentiment Classification Analysis of Tokopedia Reviews Using TF-IDF, SMOTE, and Traditional Machine Learning Models
No full textThis study explores sentiment classification on Tokopedia user reviews using TF-IDF for feature extraction and SMOTE to handle class imbalance. From nearly one million raw reviews sourced from Kaggle ("E-Commerce Ratings and Reviews in Bahasa Indonesia"), a final set of 6,477 relevant entries was obtained after rigorous preprocessing, including case folding, noise removal (emojis, URLs, numbers), normalization to KBBI standards, tokenization, stopword removal, and stemming with Sastrawi. The dataset consisted of 5,213 positive and 1,264 negative reviews (80.4% positive). SMOTE balanced the classes to 10,426 reviews with a 1:1 ratio for training. Five traditional machine learning models were evaluated: Naive Bayes, Logistic Regression, Support Vector Machine (SVM), Decision Tree, and Random Forest. Assessments were based on accuracy, precision, recall, F1-score, ROC-AUC, and computational time, using an 80:20 stratified split and 5-fold cross-validation. Random Forest achieved the best overall performance (accuracy: 0.9163, F1-score: 0.9133, ROC-AUC: 0.9784), while tuned SVM (C=10, RBF kernel) attained the highest accuracy of 0.9473 and F1-score of 0.9321. Cross-validation on Naive Bayes showed consistent results with an average accuracy of 88.09%. Further analysis using Logistic Regression coefficients identified influential features: positive sentiment associated with words like "mantap", "mudah", and "sukses", while negative sentiment correlated with "kecewa", "parah", and "lemot". These insights provide practical value for Tokopedia\u27s teams to enhance user experience, such as improving app speed and addressing complaints. The findings demonstrate the effectiveness and efficiency of traditional machine learning techniques for sentiment analysis in Bahasa Indonesia contexts.This study explores sentiment classification on Tokopedia user reviews using TF-IDF for feature extraction and SMOTE to handle class imbalance. From nearly one million raw reviews sourced from Kaggle ("E-Commerce Ratings and Reviews in Bahasa Indonesia"), a final set of 6,477 relevant entries was obtained after rigorous preprocessing, including case folding, noise removal (emojis, URLs, numbers), normalization to KBBI standards, tokenization, stopword removal, and stemming with Sastrawi. The dataset consisted of 5,213 positive and 1,264 negative reviews (80.4% positive). SMOTE balanced the classes to 10,426 reviews with a 1:1 ratio for training. Five traditional machine learning models were evaluated: Naive Bayes, Logistic Regression, Support Vector Machine (SVM), Decision Tree, and Random Forest. Assessments were based on accuracy, precision, recall, F1-score, ROC-AUC, and computational time, using an 80:20 stratified split and 5-fold cross-validation. Random Forest achieved the best overall performance (accuracy: 0.9163, F1-score: 0.9133, ROC-AUC: 0.9784), while tuned SVM (C=10, RBF kernel) attained the highest accuracy of 0.9473 and F1-score of 0.9321. Cross-validation on Naive Bayes showed consistent results with an average accuracy of 88.09%. Further analysis using Logistic Regression coefficients identified influential features: positive sentiment associated with words like "mantap", "mudah", and "sukses", while negative sentiment correlated with "kecewa", "parah", and "lemot". These insights provide practical value for Tokopedia\u27s teams to enhance user experience, such as improving app speed and addressing complaints. The findings demonstrate the effectiveness and efficiency of traditional machine learning techniques for sentiment analysis in Bahasa Indonesia contexts
Optimizing Support Vector Machine (SVM) for Sentiment Analysis of Blu by BCA Reviews with Chi-Square
No full textOne of the products resulting from the development of financial technology is the blu by BCA application. This app can be downloaded by BCA bank users via the Google Play Store and has received various user responses in the form of reviews. Analyzing these user reviews can serve as a valuable reference for further development and decision-making by BCA regarding the blu app. Sentiment analysis is conducted using the Support Vector Machine (SVM) algorithm, with SMOTE and TF-IDF techniques, and feature selection via Chi-Square. Sentiment classification using the SVM algorithm and feature selection has produced various outcomes in previous studies. Therefore, further research is necessary to analyze reviews of the blu application. This study aims to optimize the SVM method in analyzing user sentiment on the blu by BCA application by applying Chi-Square feature selection to improve sentiment classification performance. The research method includes the following stages: scraping, preprocessing, labeling, TF-IDF transformation, Chi-Square feature selection, SMOTE, data splitting, data mining, and evaluation. The testing results show that the RBF kernel achieved the highest performance with an accuracy of 0.8623, precision of 0.8623, recall of 0.8623, and F1-score of 0.8623. After applying Chi-Square feature selection, the accuracy improved to 0.8726, with precision of 0.8747, recall of 0.8725, and F1-score of 0.8723. This optimization successfully increased the accuracy by 0.0103 or 1.03%, while also improving precision, recall, and F1-score, indicating that feature selection contributes significantly to sentiment classification performance.One of the products resulting from the development of financial technology is the blu by BCA application. This app can be downloaded by BCA bank users via the Google Play Store and has received various user responses in the form of reviews. Analyzing these user reviews can serve as a valuable reference for further development and decision-making by BCA regarding the blu app. Sentiment analysis is conducted using the Support Vector Machine (SVM) algorithm, with SMOTE and TF-IDF techniques, and feature selection via Chi-Square. Sentiment classification using the SVM algorithm and feature selection has produced various outcomes in previous studies. Therefore, further research is necessary to analyze reviews of the blu application. This study aims to optimize the SVM method in analyzing user sentiment on the blu by BCA application by applying Chi-Square feature selection to improve sentiment classification performance. The research method includes the following stages: scraping, preprocessing, labeling, TF-IDF transformation, Chi-Square feature selection, SMOTE, data splitting, data mining, and evaluation. The testing results show that the RBF kernel achieved the highest performance with an accuracy of 0.8623, precision of 0.8623, recall of 0.8623, and F1-score of 0.8623. After applying Chi-Square feature selection, the accuracy improved to 0.8726, with precision of 0.8747, recall of 0.8725, and F1-score of 0.8723. This optimization successfully increased the accuracy by 0.0103 or 1.03%, while also improving precision, recall, and F1-score, indicating that feature selection contributes significantly to sentiment classification performance
Aspect-Based Sentiment Analysis of Hospital Service Reviews Using Fine-Tuned IndoBERT
No full textAspect-Based Sentiment Analysis (ABSA) has become a crucial approach for extracting detailed opinions from user-generated content, especially in the healthcare domain. This study analyzes public sentiment toward hospital services in Indonesia using IndoBERT, fine-tuned on 2.448 reviews collected from Google Reviews and Instagram. Sentiment labels were automatically assigned with a pre-trained Indonesian RoBERTa classifier, while aspect extraction was performed through a lexicon-based approach covering five service dimensions: Facilities, Staff Competence, Empathy and Communication, Reliability and Responsiveness, and Cost and Affordability. To address class imbalance, the IndoBERT model was optimized using class weight adjustments. The results demonstrate strong performance, achieving an overall accuracy of 96%. In terms of sentiment classification, the model obtained F1-scores of 89% for negative, 83% for neutral, and 99% for positive sentiment, with a macro-average F1 of 90%. By aspect, Facilities (82.24%) and Empathy & Communication (91.71%) received the highest positive sentiment, while Cost & Affordability recorded the highest proportion of negative sentiment (25%). These findings underscore the effectiveness of IndoBERT-based ABSA in capturing nuanced public perceptions and highlight its potential as a decision-support tool for hospitals to enhance service quality and patient satisfaction in Indonesia
Sentiment Analysis customer Towards Cinema Services in Semarang Using Naive Bayes Classifier on Google Reviews
No full textThe development of the entertainment industry, especially in the field of cinema, encourages every service provider to continuously maintain the quality of their services. One method of assessing customer satisfaction is through sentiment testing. The main objective of this study is to examine customer sentiment towards cinema services in Semarang by applying the Naive Bayes Classifier method. The research data was taken from 600 customer reviews on Google Review, which were then divided into two groups: training data consisting of 480 reviews (80%) and testing data consisting of 120 reviews (20%). Before the classification process, the data underwent pre-processing stages involving data cleaning, case folding, tokenization, stopword removal, and stemming, followed by data labeling into two sentiment categories, namely positive and negative. This study took five cinemas as objects, namely CitraXXI, Cinépolis Java Mall, Paragon XXI, XXI Uptown Mall, and XXI DP Mall. The classification results show that the Naive Bayes algorithm is able to group sentiments quite well, with model accuracy ranging from 0.90 to 0.94. Of the five cinemas, Cinépolis Java Mall achieved the highest accuracy, which was 0.94.The development of the entertainment industry, especially in the field of cinema, encourages every service provider to continuously maintain the quality of their services. One method of assessing customer satisfaction is through sentiment testing. The main objective of this study is to examine customer sentiment towards cinema services in Semarang by applying the Naive Bayes Classifier method. The research data was taken from 600 customer reviews on Google Review, which were then divided into two groups: training data consisting of 480 reviews (80%) and testing data consisting of 120 reviews (20%). Before the classification process, the data underwent pre-processing stages involving data cleaning, case folding, tokenization, stopword removal, and stemming, followed by data labeling into two sentiment categories, namely positive and negative. This study took five cinemas as objects, namely CitraXXI, Cinépolis Java Mall, Paragon XXI, XXI Uptown Mall, and XXI DP Mall. The classification results show that the Naive Bayes algorithm is able to group sentiments quite well, with model accuracy ranging from 0.90 to 0.94. Of the five cinemas, Cinépolis Java Mall achieved the highest accuracy, which was 0.94
Prototype of Temperature, Humidity and Fire Detection Monitoring System in Rice Warehouse Based on ESP32 Microcontroller
No full textRice warehouses in Indonesia experience significant post-harvest losses, reported to reach 10–20% annually, primarily due to poor environmental control and fire incidents. This study develops and evaluates an Internet of Things (IoT)-based environmental monitoring prototype for rice warehouses, utilizing the ESP32 microcontroller, DHT22 temperature-humidity sensor, and a flame sensor. The ESP32 was chosen for its low power consumption and robust connectivity, while DHT22 and the flame sensor were selected for their balance of accuracy, sensitivity, and cost-effectiveness. System calibration employed a digital thermohygrometer and a standard flame detector to ensure measurement validity. Experimental tests were conducted in a controlled laboratory setting with three sensor points, simulating temperature variations of 28–45°C and humidity of 60–95%, together with 24-hour reliability tests and scenarios involving fire detection at a 30 cm distance. The system achieved sensor error margins within ±0.5°C for temperature and ±2% for humidity, with actuator response times of 1–3 seconds. Real-time Telegram notifications were successfully delivered within 2–3 seconds. The integration of multi-sensors, automated actuators, and instant notifications distinguishes the proposed system from conventional approaches and previous studies. While effective for small-to-medium scale warehouses, limitations remain in fire sensor coverage and dependence on internet connectivity. The system offers an adaptable, efficient, and reliable solution to minimize manual errors and improve rice warehouse management. Future work will address broader scalability, additional gas sensors, GSM communication, and cloud-based data logging for enhanced safety and analytics.Rice warehouses in Indonesia experience significant post-harvest losses, reported to reach 10–20% annually, primarily due to poor environmental control and fire incidents. This study develops and evaluates an Internet of Things (IoT)-based environmental monitoring prototype for rice warehouses, utilizing the ESP32 microcontroller, DHT22 temperature-humidity sensor, and a flame sensor. The ESP32 was chosen for its low power consumption and robust connectivity, while DHT22 and the flame sensor were selected for their balance of accuracy, sensitivity, and cost-effectiveness. System calibration employed a digital thermohygrometer and a standard flame detector to ensure measurement validity. Experimental tests were conducted in a controlled laboratory setting with three sensor points, simulating temperature variations of 28–45°C and humidity of 60–95%, together with 24-hour reliability tests and scenarios involving fire detection at a 30 cm distance. The system achieved sensor error margins within ±0.5°C for temperature and ±2% for humidity, with actuator response times of 1–3 seconds. Real-time Telegram notifications were successfully delivered within 2–3 seconds. The integration of multi-sensors, automated actuators, and instant notifications distinguishes the proposed system from conventional approaches and previous studies. While effective for small-to-medium scale warehouses, limitations remain in fire sensor coverage and dependence on internet connectivity. The system offers an adaptable, efficient, and reliable solution to minimize manual errors and improve rice warehouse management. Future work will address broader scalability, additional gas sensors, GSM communication, and cloud-based data logging for enhanced safety and analytics
Efficient Feature Extraction Using MobileNetV2 and EfficientNetB0 for Multi-Class Brain Tumor Classification
No full textBrain tumor classification in MRI is complicated by the similarity of imaging features across multiple tumor classes. This study evaluates the use of lightweight convolutional neural network (CNN) architectures as feature extractors combined with machine learning classifiers for multi-class classification. MobileNetV2 and EfficientNetB0 were used to extract fixed-length feature representations, which were then classified using Support Vector Machine (SVM), Logistic Regression, Random Forest, and K-Nearest Neighbors. The evaluation used stratified five-fold cross-validation, and performance was measured with accuracy, F1-score, and Matthews Correlation Coefficient (MCC). Results show that EfficientNetB0 features paired with SVM achieved the highest test accuracy (98.5%), while Logistic Regression also yielded competitive performance (97.1%). Class-wise analysis indicated strong results for pituitary and non-tumor cases. This work shows that lightweight CNN-based feature extraction may serve as a practical direction for improving multi-class brain tumor MRI classification, with potential benefits for applications in resource-limited environments
Analysis of Deep Learning Algorithms Using ConvNeXt and Vision Transformer for Brain Tumor Disease
No full textThis study aims to conduct a comparative analysis and identify the most effective deep learning architecture between ConvNeXt and Vision Transformer (ViT) for the automated classification of brain tumors from MRI imagery. Rapid and accurate brain tumor diagnosis is crucial; however, the manual interpretation of MRI scans is time-consuming and reliant on specialist expertise, creating an urgent need for reliable automation in brain tumor diagnosis. This research utilizes a dataset of 4,600 images, balanced between 2,513 \u27Brain Tumor\u27 and 2,087 \u27Healthy\u27 instances. A robust 5-Fold Cross-Validation methodology was employed to evaluate model performance, wherein the data was divided into five folds, each consisting of 920 images, ensuring every image served as both training and testing data. The quantitative results demonstrated high efficacy from both models, although ConvNeXt achieved a slight, consistent advantage. ConvNeXt obtained an accuracy of 99.13%, precision of 99.13%, recall of 99.13%, and an F1-Score of 99.13%. In comparison, the ViT model scored an accuracy of 98.13%, precision of 98.14%, recall of 98.13%, and an F1-Score of 98.13%. This quantitative superiority was validated through qualitative analysis using saliency maps, which confirmed that the models\u27 computational attention was accurately focused on the anatomical locations of the actual tumor lesions
Analysis of Naive Bayes Algorithm for Lung Cancer Risk Prediction Based on Lifestyle Factors
No full textLung cancer is one of the types of cancer with the highest mortality rate in the world, which is often difficult to detect in the early stages due to minimal symptoms. This study aims to build a lung cancer risk prediction model based on lifestyle factors using the Gaussian Naive Bayes algorithm. Data fit is addressed using the Synthetic Minority Over-sampling Technique (SMOTE), and feature selection is carried out using the Mutual Information. The dataset used consists of 1000 patient data with 24 features related to lifestyle and environmental factors. Model validation is carried out using 5-fold Stratified Cross Validation, and evaluated based on accuracy, precision, recall, and confusion matrices. The results show that the application of SMOTE successfully increases the model accuracy to 91.00% with high precision and recall values in all risk classes (Low, Medium, High). The features "Passive Smoker" and "Coughing up Blood" are identified as the most influential factors in the prediction. The results of this study indicate that the combination of Gaussian Naive Bayes with SMOTE and Mutual Information is able to produce an accurate prediction model.Lung cancer is one of the types of cancer with the highest mortality rate in the world, which is often difficult to detect in the early stages due to minimal symptoms. This study aims to build a lung cancer risk prediction model based on lifestyle factors using the Gaussian Naive Bayes algorithm. Data fit is addressed using the Synthetic Minority Over-sampling Technique (SMOTE), and feature selection is carried out using the Mutual Information. The dataset used consists of 1000 patient data with 24 features related to lifestyle and environmental factors. Model validation is carried out using 5-fold Stratified Cross Validation, and evaluated based on accuracy, precision, recall, and confusion matrices. The results show that the application of SMOTE successfully increases the model accuracy to 91.00% with high precision and recall values in all risk classes (Low, Medium, High). The features "Passive Smoker" and "Coughing up Blood" are identified as the most influential factors in the prediction. The results of this study indicate that the combination of Gaussian Naive Bayes with SMOTE and Mutual Information is able to produce an accurate prediction model
Improving News Text Classification Using a Hybrid C5.0-KNN Model
No full textIn the digital era, the overwhelming volume of online news far exceeds readers’ ability to manually filter information, necessitating automated text classification. However, achieving high classification accuracy remains challenging, especially in low-resource languages like IndonesianThe C5.0 decision tree and K-Nearest Neighbors (KNN) offer complementary strengths but have not yet been jointly utilized for Indonesian news classification; therefore, this study proposes a hybrid C5.0–KNN model designed to enhance news classification performance. A dataset of 1.700 articles was collected from four Indonesian online news, namely CNN Indonesia, Okezone, Tribun Jakarta, and Tribun Jabar, covering five topical categories, namely economy/ekonomi, technology/teknologi, sport/olahraga, entertainment/hiburan, or life style/gaya hidup). The data underwent preprocessing and TF-IDF weighing before classification with the hybrid model. In this approach, C5.0 first generates interpretable decision rules, and KNN then refines borderline cases, combining rule-based and instance-based methods. The findings revealed that the hybrid model achieved a highest accuracy of 0.8847 (using 25% test data and k=5), outperforming standalone C5.0 (0.7426) and KNN (0.8735). Notably, it attained 100% recall for “sport/olahraga” and an F1-score of 0.89 for “entertainment/hiburan”. These results demonstrate the model’s novelty, efficiency, and strong potential for real-world news classification in low-resource language contexts, offering practical value for journalists, analysts, and media monitoring systems