Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Smart Valve Irrigation System Using Fuzzy Logic for Mustard
No full textThis study presents the design and implementation of a smart irrigation system using Mamdani fuzzy logic integrated with IoT-based environmental sensors. The system utilizes an ESP32 microcontroller, DHT22 temperature sensor, capacitive soil moisture sensor, and a solenoid valve to perform adaptive irrigation based on real-time environmental conditions. The fuzzy logic engine processes sensor inputs and determines the irrigation intensity through centroid-based defuzzification. A web-based dashboard was developed using PHP and JavaScript to monitor temperature, soil moisture, and irrigation status in real time. The system was tested on mustard greens (Brassica juncea L.) for 12 hours, resulting in a 35% water usage reduction compared to manual watering methods while maintaining optimal soil moisture. This approach demonstrates a promising solution for sustainable and efficient smart agriculture
A Conceptual Hybrid AI-Cloud Model for Government Information Systems: A Structured Literature Review
No full textThis study develops a comprehensive Hybrid AI-Cloud conceptual model to enhance government information systems through digital transformation. Using a systematic literature review (PRISMA protocol) of 51 publications (2020-2025) from Scopus, IEEE Xplore, and ScienceDirect, we identify four critical components: a hybrid architecture combining private and public clouds achieves 97.46% prediction accuracy but faces interoperability challenges in Indonesia where 85% of agencies use disparate systems; layered security with Hyperledger Fabric blockchain reduces data breaches by 72%, though 65% of Indonesian institutions lack CSIRT teams; user-centric designs score 76.88 on SUS scales yet encounter 71% civil servant resistance to AI automation; and organizational adoption strategies based on UTAUT frameworks are hindered by only 12% of civil servants having digital certifications. The research reveals Indonesia\u27s significant gaps in system integration, cybersecurity preparedness, and digital literacy compared to global leaders like Estonia and Singapore. Successful implementation requires standardized cloud architectures with API gateways, mandatory cybersecurity audits, comprehensive digital training programs, and phased adoption roadmaps with change management components. While offering a holistic framework for digital government transformation, the study acknowledges limitations including literature bias toward developed nations and the need for local empirical validation through pilot projects, suggesting future research should incorporate ethical AI governance considerations alongside technical implementations.This study develops a comprehensive Hybrid AI-Cloud conceptual model to enhance government information systems through digital transformation. Using a systematic literature review (PRISMA protocol) of 51 publications (2020-2025) from Scopus, IEEE Xplore, and ScienceDirect, we identify four critical components: a hybrid architecture combining private and public clouds achieves 97.46% prediction accuracy but faces interoperability challenges in Indonesia where 85% of agencies use disparate systems; layered security with Hyperledger Fabric blockchain reduces data breaches by 72%, though 65% of Indonesian institutions lack CSIRT teams; user-centric designs score 76.88 on SUS scales yet encounter 71% civil servant resistance to AI automation; and organizational adoption strategies based on UTAUT frameworks are hindered by only 12% of civil servants having digital certifications. The research reveals Indonesia\u27s significant gaps in system integration, cybersecurity preparedness, and digital literacy compared to global leaders like Estonia and Singapore. Successful implementation requires standardized cloud architectures with API gateways, mandatory cybersecurity audits, comprehensive digital training programs, and phased adoption roadmaps with change management components. While offering a holistic framework for digital government transformation, the study acknowledges limitations including literature bias toward developed nations and the need for local empirical validation through pilot projects, suggesting future research should incorporate ethical AI governance considerations alongside technical implementations
Development of MobileNetV2 for CT-Scan Lung Classification Using Transfer Learning
No full textLung cancer is one of the leading causes of cancer-related deaths worldwide, making early detection crucial for improving patient survival rates. This study proposes an automated classification approach based on deep learning using the MobileNetV2 architecture to identify three categories of lung CT scan images: normal, benign, and malignant. The dataset used is the augmented IQ-OTH/NCCD Lung Cancer Dataset, consisting of 3,609 images with a resolution of 224×224 pixels. All images underwent preprocessing steps including RGB conversion, pixel rescaling, and normalization. The MobileNetV2 model was modified by adding a GlobalAveragePooling2D layer, a dense layer, and dropout to reduce overfitting risk. Training was conducted for 28 epochs using the optimizer Adam, followed by evaluation using accuracy, precision, recall, and F1-score metrics. The model was tested on unseen data and validated using Stratified 5-Fold Cross Validation. The testing results showed an overall accuracy of 97%, with a perfect recall score (1.00) for the malignant class. The cross-validation yielded an average accuracy of 97.26% with a standard deviation of ±0.66%, indicating consistent model performance. Given its lightweight architecture and high accuracy, MobileNetV2 has the potential to be implemented as a decision support system in medical image analysis.Lung cancer is one of the leading causes of cancer-related deaths worldwide, making early detection crucial for improving patient survival rates. This study proposes an automated classification approach based on deep learning using the MobileNetV2 architecture to identify three categories of lung CT scan images: normal, benign, and malignant. The dataset used is the augmented IQ-OTH/NCCD Lung Cancer Dataset, consisting of 3,609 images with a resolution of 224×224 pixels. All images underwent preprocessing steps including RGB conversion, pixel rescaling, and normalization. The MobileNetV2 model was modified by adding a GlobalAveragePooling2D layer, a dense layer, and dropout to reduce overfitting risk. Training was conducted for 28 epochs using the optimizer Adam, followed by evaluation using accuracy, precision, recall, and F1-score metrics. The model was tested on unseen data and validated using Stratified 5-Fold Cross Validation. The testing results showed an overall accuracy of 97%, with a perfect recall score (1.00) for the malignant class. The cross-validation yielded an average accuracy of 97.26% with a standard deviation of ±0.66%, indicating consistent model performance. Given its lightweight architecture and high accuracy, MobileNetV2 has the potential to be implemented as a decision support system in medical image analysis
Development of an IoT-Based Smart Greenhouse with Fuzzy Logic for Chrysanthemum Cultivation
No full textConventional cultivation of Chrysanthemum plants in greenhouses faces serious challenges such as inefficiency, response delays, and errors in temperature and humidity settings due to manual management. These conditions result in unsuitable growing environments that can reduce the quality and quantity of harvests. To overcome these problems, this study developed a smart greenhouse system based on the Internet of Things (IoT) and cloud computing with the application of fuzzy logic. The system is designed to automatically monitor and control temperature, humidity, and light intensity using NodeMCU ESP32, DHT22 and BH1750 sensors, as well as relay-based actuators and mini air conditioners. Environmental data is sent to the cloud and processed using the Sugeno fuzzy method to produce adaptive and precise control decisions. Test results show that the system can maintain stable and optimal environmental conditions with an average temperature control difference of 30.341% and an actuator efficiency of 9.34% against microcontroller commands. This system provides a modern solution to the limitations of traditional methods, and supports smart agriculture in tropical climates such as Lhokseumawe
A Hybrid Approach to Music Recommendations Based on Audio Similarity Using Autoencoder and LightGBM
No full textMusic recommendation systems help users navigate large music collections by suggesting songs aligned with their preferences. However, conventional methods often overlook the depth of audio content, limiting personalization and accuracy. This study proposes a hybrid approach that uses PCA and Autoencoder to extract audio embeddings. These embeddings are processed using K-Nearest Neighbors to find similar tracks, followed by a reranking step with LightGBM based on predicted relevance. The system achieved strong results: 98% accuracy, 0.96 precision, 0.96 recall, and 0.96 F1-score for the Similar class, with 0.99 precision and recall for Not Similar. Cross-validation confirmed model robustness, with an average accuracy of 97.99%, precision of 0.9577, recall of 0.9624, and F1-score of 0.9600, all with low standard deviations. These outcomes show that combining deep audio features with machine learning ranking enhances recommendation quality. Future improvements may involve incorporating metadata and genre-based visualizations for more diverse and interpretable results.Music recommendation systems help users navigate large music collections by suggesting songs aligned with their preferences. However, conventional methods often overlook the depth of audio content, limiting personalization and accuracy. This study proposes a hybrid approach that uses PCA and Autoencoder to extract audio embeddings. These embeddings are processed using K-Nearest Neighbors to find similar tracks, followed by a reranking step with LightGBM based on predicted relevance. The system achieved strong results: 98% accuracy, 0.96 precision, 0.96 recall, and 0.96 F1-score for the Similar class, with 0.99 precision and recall for Not Similar. Cross-validation confirmed model robustness, with an average accuracy of 97.99%, precision of 0.9577, recall of 0.9624, and F1-score of 0.9600, all with low standard deviations. These outcomes show that combining deep audio features with machine learning ranking enhances recommendation quality. Future improvements may involve incorporating metadata and genre-based visualizations for more diverse and interpretable results
Implementation of the Support Vector Machine (SVM) Method for Classifying the Maturity Level of Oil Palm Fruit
No full textThis study discusses the classification of palm fruit ripeness levels using the Support Vector Machine (SVM) method. Palm fruit ripeness significantly affects the yield and quality of the oil produced. By utilizing image processing techniques, colour and texture features are extracted from the fruit images to support the classification process. The SVM model was trained with a dataset covering various ripeness levels, including unripe, ripe, overripe, and rotten. The evaluation results show the high accuracy of the SVM model in identifying ripeness levels. This study highlights the potential of machine learning technology in improving the productivity and quality of agricultural products. Support Vector Machine (SVM) is a machine learning method used to classify data into categories by finding the optimal dividing line between two classes, thereby maximizing the distance between the data from the two classes. SVM itself has proven to be very effective in detecting images, as evidenced by several studies such as detecting the ripeness level of melon fruit, each producing a model with an accuracy level above 86%. Thus, this study uses SVM suitable for use in detecting the ripeness level of oil palm fruit. This study produced an SVM model with an accuracy level of 93%.This study discusses the classification of palm fruit ripeness levels using the Support Vector Machine (SVM) method. Palm fruit ripeness significantly affects the yield and quality of the oil produced. By utilizing image processing techniques, colour and texture features are extracted from the fruit images to support the classification process. The SVM model was trained with a dataset covering various ripeness levels, including unripe, ripe, overripe, and rotten. The evaluation results show the high accuracy of the SVM model in identifying ripeness levels. This study highlights the potential of machine learning technology in improving the productivity and quality of agricultural products. Support Vector Machine (SVM) is a machine learning method used to classify data into categories by finding the optimal dividing line between two classes, thereby maximizing the distance between the data from the two classes. SVM itself has proven to be very effective in detecting images, as evidenced by several studies such as detecting the ripeness level of melon fruit, each producing a model with an accuracy level above 86%. Thus, this study uses SVM suitable for use in detecting the ripeness level of oil palm fruit. This study produced an SVM model with an accuracy level of 93%
Hyperparameter Optimization and Feature Selection Analysis on the XGBoost Model for Hepatitis C Infection Prediction
No full textHepatitis C is a liver disease that can progress to chronic conditions such as cirrhosis and liver cancer. Early detection is essential and can be supported through machine learning approaches. This study analyzes the effect of feature selection and hyperparameter tuning on the performance of the XGBoost model in classifying hepatitis C infection. The dataset, obtained from Kaggle, contains laboratory test attributes. The preprocessing stage involved handling missing values, encoding categorical variables, removing outlier classes, and normalizing data using StandardScaler. After stratified splitting, the training set was balanced using the SMOTE technique. Feature selection was carried out using the ANOVA F-score method, and hyperparameter tuning was performed using GridSearchCV. Three model scenarios were compared: baseline, with feature selection, and with combined feature selection and hyperparameter tuning. The evaluation results showed that the third model achieved the best performance with 96% accuracy, 79% precision, 81% recall, and a 78% F1-score, despite a slight decrease in the ROC AUC value. This approach has proven effective in improving model performance and is relevant for supporting more accurate hepatitis C diagnosis systems.Hepatitis C is a liver disease that can progress to chronic conditions such as cirrhosis and liver cancer. Early detection is essential and can be supported through machine learning approaches. This study analyzes the effect of feature selection and hyperparameter tuning on the performance of the XGBoost model in classifying hepatitis C infection. The dataset, obtained from Kaggle, contains laboratory test attributes. The preprocessing stage involved handling missing values, encoding categorical variables, removing outlier classes, and normalizing data using StandardScaler. After stratified splitting, the training set was balanced using the SMOTE technique. Feature selection was carried out using the ANOVA F-score method, and hyperparameter tuning was performed using GridSearchCV. Three model scenarios were compared: baseline, with feature selection, and with combined feature selection and hyperparameter tuning. The evaluation results showed that the third model achieved the best performance with 96% accuracy, 79% precision, 81% recall, and a 78% F1-score, despite a slight decrease in the ROC AUC value. This approach has proven effective in improving model performance and is relevant for supporting more accurate hepatitis C diagnosis systems
Programming Assessment in E-Learning through Rule-Based Automatic Question Generation with Large Language Models
No full textThis study develops an evaluation instrument for Python programming using a Rule-Based Automatic Question Generation (AQG) system integrated with Large Language Models (LLMs), designed based on the Revised Bloom’s Taxonomy. The urgency of this research stems from the limitations of conventional programming evaluations, which are often time-consuming, less objective, and insufficiently aligned with cognitive learning levels. The proposed method applies assessment terms as rule-based constraints to guide LLM-generated questions, ensuring both pedagogical validity and structural consistency in JSON format. A total of 91 questions were produced, consisting of multiple-choice and coding items, which were then validated by three programming experts and tested on 32 vocational students. The findings indicate that the instrument achieved an overall validity of 77.66% (valid category), with the highest accuracy at the Apply (96.30%) and Create (100%) levels. The reliability test using Cronbach’s Alpha yielded 0.721, showing acceptable internal consistency. Item difficulty analysis revealed a strong dominance of easy questions (97.78%), with only 2.22% classified as moderate and none as difficult. Student performance also showed a fluctuating pattern: high in Remember (94.79%), Understand (95.83%), and Create (95.60%), but lower in Apply (86.11%), Analyze (90.97%), and Evaluate (87.15%). These results confirm that integrating Rule-Based AQG with LLMs can produce valid, reliable, and adaptive evaluation instruments that not only capture basic programming competencies but also partially address higher-order cognitive skills. This research contributes both practically by providing educators with an efficient tool for generating evaluation items and academically by enriching the growing body of literature on AI-assisted assessment in programming education.This study develops an evaluation instrument for Python programming using a Rule-Based Automatic Question Generation (AQG) system integrated with Large Language Models (LLMs), designed based on the Revised Bloom’s Taxonomy. The urgency of this research stems from the limitations of conventional programming evaluations, which are often time-consuming, less objective, and insufficiently aligned with cognitive learning levels. The proposed method applies assessment terms as rule-based constraints to guide LLM-generated questions, ensuring both pedagogical validity and structural consistency in JSON format. A total of 91 questions were produced, consisting of multiple-choice and coding items, which were then validated by three programming experts and tested on 32 vocational students. The findings indicate that the instrument achieved an overall validity of 77.66% (valid category), with the highest accuracy at the Apply (96.30%) and Create (100%) levels. The reliability test using Cronbach’s Alpha yielded 0.721, showing acceptable internal consistency. Item difficulty analysis revealed a strong dominance of easy questions (97.78%), with only 2.22% classified as moderate and none as difficult. Student performance also showed a fluctuating pattern: high in Remember (94.79%), Understand (95.83%), and Create (95.60%), but lower in Apply (86.11%), Analyze (90.97%), and Evaluate (87.15%). These results confirm that integrating Rule-Based AQG with LLMs can produce valid, reliable, and adaptive evaluation instruments that not only capture basic programming competencies but also partially address higher-order cognitive skills. This research contributes both practically by providing educators with an efficient tool for generating evaluation items and academically by enriching the growing body of literature on AI-assisted assessment in programming education
Multiclass Classification of Tomato Leaf Diseases Using GLCM, Color, and Shape Feature Extraction with Optimized XGBoost
No full textAutomatic classification of tomato leaf diseases is an essential component in advancing precision agriculture based on artificial intelligence. This study aims to develop a multiclass classification model for tomato leaf diseases by utilizing texture, color, and shape features, and employing an optimized XGBoost algorithm. The public PlantVillage dataset was used, with preprocessing stages including feature extraction, normalization, dimensionality reduction using PCA, and class balancing using SMOTE. The experimental results showed that the model successfully classified ten disease classes with a high accuracy of 97.63%, and both macro and weighted f1-scores of 0.98. These findings indicate that the combination of handcrafted features and XGBoost offers an effective, efficient, and applicable solution for plant disease diagnostic systems.Automatic classification of tomato leaf diseases is an essential component in advancing precision agriculture based on artificial intelligence. This study aims to develop a multiclass classification model for tomato leaf diseases by utilizing texture, color, and shape features, and employing an optimized XGBoost algorithm. The public PlantVillage dataset was used, with preprocessing stages including feature extraction, normalization, dimensionality reduction using PCA, and class balancing using SMOTE. The experimental results showed that the model successfully classified ten disease classes with a high accuracy of 97.63%, and both macro and weighted f1-scores of 0.98. These findings indicate that the combination of handcrafted features and XGBoost offers an effective, efficient, and applicable solution for plant disease diagnostic systems
Enhancing Aspect-Based Sentiment Analysis via Hugging Face Fine-Tuned IndoBERT
No full textAspect-Based Sentiment Analysis (ABSA) on hotel reviews faces significant challenges regarding semantic complexity and severe class imbalance, particularly in low-resource languages like Indonesian. This study evaluates the effectiveness of fine-tuning IndoBERT, a pre-trained Transformer model, to address these issues by benchmarking it against classical statistical methods (TF-IDF) and static embeddings (Sentence-BERT). Utilizing the HoASA dataset, the experiment implements a Random Oversampling strategy at the text level to mitigate data sparsity in minority classes. Empirical results demonstrate that the fine-tuned IndoBERT significantly outperforms baselines on the majority of aspects, achieving a global accuracy of 97% and macro F1-score of 0.92. Granular per-aspect analysis reveals that the model’s self-attention mechanism captures linguistic context robustly in tangible aspects (e.g., wifi, service), yet faces persistent challenges in highly ambiguous aspects such as smell (bau) and general. Statistical significance tests (Paired t-test and Wilcoxon) confirm that the performance gains over baselines are statistically significant (p < 0.05) and not due to random chance. The study concludes that leveraging contextual representations from IndoBERT, combined with data balancing strategies, offers a superior and statistically robust solution for handling linguistic variations and class bias in the Indonesian hospitality domain.Aspect-Based Sentiment Analysis (ABSA) on hotel reviews faces significant challenges regarding semantic complexity and severe class imbalance, particularly in low-resource languages like Indonesian. This study evaluates the effectiveness of fine-tuning IndoBERT, a pre-trained Transformer model, to address these issues by benchmarking it against classical statistical methods (TF-IDF) and static embeddings (Sentence-BERT). Utilizing the HoASA dataset, the experiment implements a Random Oversampling strategy at the text level to mitigate data sparsity in minority classes. Empirical results demonstrate that the fine-tuned IndoBERT significantly outperforms baselines on the majority of aspects, achieving a global accuracy of 97% and macro F1-score of 0.92. Granular per-aspect analysis reveals that the model’s self-attention mechanism captures linguistic context robustly in tangible aspects (e.g., wifi, service), yet faces persistent challenges in highly ambiguous aspects such as smell (bau) and general. Statistical significance tests (Paired t-test and Wilcoxon) confirm that the performance gains over baselines are statistically significant (p < 0.05) and not due to random chance. The study concludes that leveraging contextual representations from IndoBERT, combined with data balancing strategies, offers a superior and statistically robust solution for handling linguistic variations and class bias in the Indonesian hospitality domain