Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Comparison of Transfer learning Models MobileNetV3-Large and EfficientNet-B0 for Rice Leaf Disease Classification
No full textRice productivity strongly depends on early detection of leaf diseases, while manual identification is often delayed and subjective. This study investigates the use of lightweight CNN architectures MobileNetV3-Large and EfficientNet-B0 based on transfer learning to classify six rice leaf disease classes, namely bacterial leaf blight, brown spot, healthy, leaf blast, leaf scald, and narrow brown spot. The dataset is obtained from Kaggle and consists of 2,628 images with a balanced class distribution, stratified into training, validation, and test sets with a ratio of 80%:10%:10%. The images are resized to 224×224 pixels and data augmentation was applied to the training set. Pretrained ImageNet weights are first used as frozen feature extractors, followed by partial fine-tuning of the last 30% backbone layers, with custom classification layers trained using the Adam optimizer with an early stopping mechanism. Model performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrices, while computational efficiency is assessed based on parameter count and inference speed measured in frames per second. The results show that under partial fine-tuning MobileNetV3-Large achieves 95.83% test accuracy and 95.80% macro F1-score with 3.12 million parameters, while EfficientNet-B0 obtains 93.18% accuracy and 93.02% macro F1-score with 4.21 million parameters. Both models achieve inference speeds above 50 frames per second, suggesting their potential suitability for deployment on resource-constrained devices. Bootstrap analysis suggests the performance gap is clear in the frozen stage but becomes less conclusive after partial fine-tuning. Overall, MobileNetV3-Large provides the best trade-off between accuracy and efficiency for rice leaf disease classification.Rice productivity strongly depends on early detection of leaf diseases, while manual identification is often delayed and subjective. This study investigates the use of lightweight CNN architectures MobileNetV3-Large and EfficientNet-B0 based on transfer learning to classify six rice leaf disease classes, namely bacterial leaf blight, brown spot, healthy, leaf blast, leaf scald, and narrow brown spot. The dataset is obtained from Kaggle and consists of 2,628 images with a balanced class distribution, stratified into training, validation, and test sets with a ratio of 80%:10%:10%. The images are resized to 224×224 pixels and data augmentation was applied to the training set. Pretrained ImageNet weights are first used as frozen feature extractors, followed by partial fine-tuning of the last 30% backbone layers, with custom classification layers trained using the Adam optimizer with an early stopping mechanism. Model performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrices, while computational efficiency is assessed based on parameter count and inference speed measured in frames per second. The results show that under partial fine-tuning MobileNetV3-Large achieves 95.83% test accuracy and 95.80% macro F1-score with 3.12 million parameters, while EfficientNet-B0 obtains 93.18% accuracy and 93.02% macro F1-score with 4.21 million parameters. Both models achieve inference speeds above 50 frames per second, suggesting their potential suitability for deployment on resource-constrained devices. Bootstrap analysis suggests the performance gap is clear in the frozen stage but becomes less conclusive after partial fine-tuning. Overall, MobileNetV3-Large provides the best trade-off between accuracy and efficiency for rice leaf disease classification
Indonesian Gold Price Forecasting Using Simple and Stacked LSTM with Expanding Window
No full textThis study investigates the performance of two deep learning architectures, namely Simple LSTM and Stacked LSTM, for Indonesian gold price forecasting, with a particular focus on evaluating the effect of optimizer selection and learning rate configurations. An experimental framework is implemented using daily Indonesian gold price data from 2021 to 2024. Model performance is assessed using five-fold expanding window time series cross-validation to ensure robustness and avoid data leakage. Four adaptive training optimizers (Adam, Nadam, Adamax, and RMSprop) are evaluated across three learning-rate settings as part of a systematic sensitivity analysis of training hyperparameters. The results indicate that the Simple LSTM consistently outperforms the Stacked LSTM. The best performance is achieved by the Simple LSTM using the Adam optimizer with a learning rate of 0.01, yielding an RMSE of 9.235, MAE of 7.060, and MAPE of 0.71%. These findings demonstrate that simpler architectures combined with appropriate training configurations can provide superior forecasting accuracy for volatile financial time series
Classification Of Student Depression Using Support Vector Machine Modelling and Backward Elimination
No full textDepression among university students has become a serious mental health concern that can negatively affect academic performance and overall well-being. Early detection of Depression is essential to provide timely support and preventive interventions. This study proposes a machine learning approach to classify student Depression using a Support Vector Machine (SVM) combined with Backward Elimination (BE) for feature selection. The dataset used in this research was obtained from a public repository and consists of 502 student records with multiple psychological and demographic attributes. Data preprocessing included categorical encoding and Min–Max normalization, followed by an 80:20 split for training and testing. Experimental results show that the baseline SVM model achieved an accuracy of 0.9208, while the application of Backward Elimination improved the performance to 0.9604. In addition, precision, recall, and F1-score also showed notable improvements, indicating a reduction in misclassification, particularly for non-depressed students. These findings demonstrate that integrating feature selection with SVM can enhance classification performance and provide a more efficient model for supporting early Depression detection among university students.Depression among university students has become a serious mental health concern that can negatively affect academic performance and overall well-being. Early detection of Depression is essential to provide timely support and preventive interventions. This study proposes a machine learning approach to classify student Depression using a Support Vector Machine (SVM) combined with Backward Elimination (BE) for feature selection. The dataset used in this research was obtained from a public repository and consists of 502 student records with multiple psychological and demographic attributes. Data preprocessing included categorical encoding and Min–Max normalization, followed by an 80:20 split for training and testing. Experimental results show that the baseline SVM model achieved an accuracy of 0.9208, while the application of Backward Elimination improved the performance to 0.9604. In addition, precision, recall, and F1-score also showed notable improvements, indicating a reduction in misclassification, particularly for non-depressed students. These findings demonstrate that integrating feature selection with SVM can enhance classification performance and provide a more efficient model for supporting early Depression detection among university students
Design and Development of UI/UX HRIS Application for Human Resource Management at PT Kreatif System Indonesia
No full textThe rapid shift toward digital business requires organizations to implement integrated Human Resource Management (HRM) systems to reduce inefficiencies caused by manual processes. At PT Kreatif System Indonesia, the lack of integration between the fingerprint attendance system and core HR functions resulted in fragmented data, higher human error risk, and reduced decision-making accuracy. This study focused exclusively on PT Kreatif System Indonesia to address its internal operational challenges. The research aimed to design a consistent, easy-to-use, and memorable UI/UX based on Material Design principles and validate it through comprehensive usability evaluation. Interface consistency was emphasized to reduce cognitive load, improve recognition, and support smooth task execution across different user skill levels. The Prototyping development model enabled iterative refinements through continuous user feedback. Usability was evaluated using Task Completion Rate (TCR), Time-Based Efficiency (TBE), and the System Usability Scale (SUS). The developed HRIS successfully integrated all targeted modules. Initial testing achieved a TCR of 87.85% and a TBE of 0.0878 goals/second, which improved to 93.45% and 0.1258 goals/second after UI/UX enhancements. A SUS score of 74.42 indicates that the system is acceptable and well received. These results confirm that consistent UI/UX design significantly enhances HRIS effectiveness, efficiency, and usability.Pergeseran cepat bisnis digital mengharuskan organisasi untuk mengimplementasikan sistem Manajemen Sumber Daya Manusia terintegrasi guna mengurangi inefisiensi yang disebabkan oleh proses manual. Di PT Kreatif System Indonesia, kurangnya integrasi antara sistem absensi sidik jari dan fungsi inti SDM mengakibatkan data yang terfragmentasi, risiko kesalahan manusia yang lebih tinggi, dan akurasi pengambilan keputusan yang menurun. Studi ini berfokus secara eksklusif pada PT Kreatif System Indonesia untuk mengatasi tantangan operasional internalnya. Penelitian ini bertujuan untuk merancang UI/UX yang konsisten, mudah digunakan, dan mudah diingat berdasarkan prinsip "Material Design" dan memvalidasinya melalui evaluasi kegunaan yang komprehensif. Konsistensi antarmuka ditekankan untuk mengurangi beban kognitif, meningkatkan pengenalan, dan mendukung kelancaran pelaksanaan tugas di berbagai tingkat keterampilan pengguna. Model pengembangan Prototyping memungkinkan penyempurnaan iteratif melalui umpan balik pengguna yang berkelanjutan. Kegunaan dievaluasi menggunakan Task Completion Rate (TCR), Time-Based Efficiency (TBE), dan System Usability Scale (SUS). HRIS yang dikembangkan berhasil mengintegrasikan semua modul yang ditargetkan. Pengujian awal mencapai TCR sebesar 87,85% dan TBE sebesar 0,0878 tujuan/detik, yang meningkat menjadi 93,45% dan 0,1258 tujuan/detik setelah peningkatan UI/UX. Skor SUS sebesar 74,42 menunjukkan bahwa sistem tersebut dapat diterima dan diterima dengan baik. Hasil ini menegaskan bahwa desain UI/UX yang konsisten secara signifikan meningkatkan efektivitas, efisiensi, dan kegunaan HRIS
Unveiling the Blockchain Intention-Behavior Gap Among Young Developers
No full textThis study assesses the factors influencing blockchain technology acceptance among young developers in Batam, Indonesia, with a specific focus on comparing two distinct behaviors: using blockchain-based applications and engaging in blockchain development. Data were collected through a survey of 215 young developers and analyzed using Partial Least Squares-Structural Equation Modeling (PLS-SEM). The main outcomes reveal two fundamentally different adoption pathways. The intention to use blockchain applications is primarily driven by personal engagement and social influence, reflecting a "hype-driven" interest, and this intention strongly translates into actual usage behavior. Conversely, the model demonstrates a complete failure to explain development behavior, revealing a significant intention-behavior gap where the intention to develop shows no significant effect on actual development activities. The study concludes that for this demographic, hype-driven interest is sufficient for superficial application adoption but wholly inadequate for fostering development capabilities. Substantive adoption requires more than social trends; therefore, industry and educational focus should shift from promoting hype to enhancing technical literacy and demonstrating tangible use cases to bridge the gap from interest to competence
Improvement of User Experience Evaluation For SMEs Digital Application Using TRI, TAM, SUS Integration
No full textMicro, Small, and Medium Enterprises (MSMEs) in the service sector, particularly vehicle wash services, continue to face challenges related to queue management, service transparency, and operational efficiency, which negatively affect user experience. This study aims to develop and evaluate a mobile-based service booking and management application prototype by integrating the Design Science Research (DSR) approach with the Technology Readiness Index (TRI), Technology Acceptance Model (TAM), and System Usability Scale (SUS) as an evaluation framework. The artifact was developed through DSR stages, including problem identification, design, demonstration, and evaluation. Qualitative data were collected through interviews with MSME owners, employees, and customers and analyzed using Thematic Analysis. Quantitative evaluation involved 106 respondents to measure technology readiness, user acceptance, and usability quality, accompanied by a descriptive analysis of relationships among the constructs. The results indicate a high level of technology readiness (TRI = 3.53) and very strong user acceptance (TAM = 4.27). However, the usability score falls within the marginal acceptable category (SUS = 62.95), indicating a gap between conceptual acceptance and actual interaction quality. These findings demonstrate that integrating TRI–TAM–SUS within the DSR framework effectively identifies critical contradictions that can serve as a basis for refining UI/UX design and implementation strategies for digital applications in service-based MSMEs
Recommendation System Yogyakarta Tourism Using TF-IDF and Cosine Similarity Methods with Word Normalizer
No full textThe abundance of tourism information in Yogyakarta often overwhelms tourists due to non-standard text data. This research develops a tourism recommendation system using Content-Based Filtering by integrating TF-IDF and Cosine Similarity algorithms, enhanced with a Word Normalizer stage. The research method involves data preprocessing including case folding, filtering, stopword removal, and stemming combined with word normalization to standardize irregular spellings. Text feature representation is calculated using TF-IDF weighting, followed by measuring similarity between destinations through vector-based Cosine Similarity. The query testing of Pantai Parangtritis against Pantai Ngandong yielded the highest similarity score of 0.9397. System performance evaluation showed a Precision@5 of 0.84, Recall@5 of 0.10, and Mean Average Precision (MAP) of 0.81. In conclusion, strengthening the method with a Word Normalizer significantly improves the validity of top-ranked recommendations, enabling tourists to accurately find relevant attractions according to their preferences.The abundance of tourism information in Yogyakarta often overwhelms tourists due to non-standard text data. This research develops a tourism recommendation system using Content-Based Filtering by integrating TF-IDF and Cosine Similarity algorithms, enhanced with a Word Normalizer stage. The research method involves data preprocessing including case folding, filtering, stopword removal, and stemming combined with word normalization to standardize irregular spellings. Text feature representation is calculated using TF-IDF weighting, followed by measuring similarity between destinations through vector-based Cosine Similarity. The query testing of Pantai Parangtritis against Pantai Ngandong yielded the highest similarity score of 0.9397. System performance evaluation showed a Precision@5 of 0.84, Recall@5 of 0.10, and Mean Average Precision (MAP) of 0.81. In conclusion, strengthening the method with a Word Normalizer significantly improves the validity of top-ranked recommendations, enabling tourists to accurately find relevant attractions according to their preferences
Implementation of LSTM for Gold Price Prediction in Indonesia
No full textGold is a significant investment instrument that serves as a safe-haven asset; nevertheless, its price dynamics are inherently nonlinear and highly volatile due to the influence of various economic factors. This study aims to develop a predictive model for daily gold prices denominated in Indonesian Rupiah. The proposed methodology employs a Long Short-Term Memory (LSTM) neural network architecture. Historical gold price data covering the period from January 1, 2015, to October 1, 2025, were obtained from investing.com. The dataset underwent a preprocessing phase, which included normalization using the MinMaxScaler and the construction of input sequences with a sliding window of 60 time steps. The implemented LSTM model consists of two stacked layers, each comprising 16 units, and is equipped with a dropout rate of 0.2 as well as an early stopping mechanism to improve generalization and prevent overfitting. The evaluation results demonstrate that the proposed model achieved a Mean Absolute Percentage Error (MAPE) of 5.08% and an accuracy of 94.92%, with a Mean Squared Error (MSE) of 0.00203. Furthermore, the visualization of prediction outcomes confirms the model’s capability to effectively capture actual price fluctuations, including during periods of heightened market volatility. Overall, these findings indicate that a relatively simple LSTM architecture is effective for forecasting gold price movements in the Indonesian market. The results of this study provide a robust foundation for the future development of more sophisticated predictive systems and potential real-time applications.Gold is a significant investment instrument that serves as a safe-haven asset; nevertheless, its price dynamics are inherently nonlinear and highly volatile due to the influence of various economic factors. This study aims to develop a predictive model for daily gold prices denominated in Indonesian Rupiah. The proposed methodology employs a Long Short-Term Memory (LSTM) neural network architecture. Historical gold price data covering the period from January 1, 2015, to October 1, 2025, were obtained from investing.com. The dataset underwent a preprocessing phase, which included normalization using the MinMaxScaler and the construction of input sequences with a sliding window of 60 time steps. The implemented LSTM model consists of two stacked layers, each comprising 16 units, and is equipped with a dropout rate of 0.2 as well as an early stopping mechanism to improve generalization and prevent overfitting. The evaluation results demonstrate that the proposed model achieved a Mean Absolute Percentage Error (MAPE) of 5.08% and an accuracy of 94.92%, with a Mean Squared Error (MSE) of 0.00203. Furthermore, the visualization of prediction outcomes confirms the model’s capability to effectively capture actual price fluctuations, including during periods of heightened market volatility. Overall, these findings indicate that a relatively simple LSTM architecture is effective for forecasting gold price movements in the Indonesian market. The results of this study provide a robust foundation for the future development of more sophisticated predictive systems and potential real-time applications
Optimizing XGBoost for Heart Disease Risk Classification Using Optuna and Random Search on the Behavioral Risk Factor Surveillance System (BRFSS) 2023 Dataset
No full textHeart disease is a critical public health issue in Indonesia, contributing to approximately 1,5 million deaths annually. Although machine learning methods, particularly Extreme Gradient Boosting (XGBoost), have demonstrated strong performance in medical classification tasks, their optimization on large-scale and highly imbalanced health datasets remains underexplored. This study optimizes XGBoost for heart disease risk classification using the Behavioral Risk Factor Surveillance System (BRFSS) 2023 dataset, consisting of 290.156 samples after preprocessing. Two hyperparameter optimization approaches, Optuna and Random Search, are evaluated across three class imbalance handling techniques, namely class weighting, SMOTE, and Random Undersampling (RUS). Model evaluation focuses on AUC and recall to prioritize sensitivity in identifying individuals at risk. The results show that the OptunaRUS and RandomWeight models achieve the most stable performance, with OptunaRUS attaining an AUC of 83,06% and a recall of 75,69% on the test dataset. Feature importance analysis indicates that age range and hypertension are the most influential predictors. These findings confirm that hyperparameter optimization on large-scale health data improves model discriminative capability and generalization, while selective sampling strategies such as RUS provide more stable performance than generative methods in high-dimensional datasets.Heart disease is a critical public health issue in Indonesia, contributing to approximately 1,5 million deaths annually. Although machine learning methods, particularly Extreme Gradient Boosting (XGBoost), have demonstrated strong performance in medical classification tasks, their optimization on large-scale and highly imbalanced health datasets remains underexplored. This study optimizes XGBoost for heart disease risk classification using the Behavioral Risk Factor Surveillance System (BRFSS) 2023 dataset, consisting of 290.156 samples after preprocessing. Two hyperparameter optimization approaches, Optuna and Random Search, are evaluated across three class imbalance handling techniques, namely class weighting, SMOTE, and Random Undersampling (RUS). Model evaluation focuses on AUC and recall to prioritize sensitivity in identifying individuals at risk. The results show that the OptunaRUS and RandomWeight models achieve the most stable performance, with OptunaRUS attaining an AUC of 83,06% and a recall of 75,69% on the test dataset. Feature importance analysis indicates that age range and hypertension are the most influential predictors. These findings confirm that hyperparameter optimization on large-scale health data improves model discriminative capability and generalization, while selective sampling strategies such as RUS provide more stable performance than generative methods in high-dimensional datasets
Detection of Diabetic Retinopathy Using Hybrid InceptionResNetV2-KELM Method
No full textDiabetic Retinopathy (DR) is a complication of Diabetes Mellitus (DM), both type 1 and type 2 DM. Based on its severity, DR is divided into mild DR, moderate DR, severe DR, and proliferative DR stages. Manual detection is difficult because there is a fairly small difference between normal and DR. The Computer-Aided Diagnosis (CAD) system is a solution for detecting the severity of DR quickly and accurately so that DR sufferers do not get worse, which can cause blindness. This study uses fundus images from the Mesindor dataset consisting of four classes, namely normal, mild DR, moderate DR, and severe DR, with the InceptionResNetV2-KELM hybrid method. InceptionResNetV2 is used as a feature extraction and Kernel Extreme Learning Machine (KELM) as its classification. Several types of kernels are applied as model trials. The results show the highest sensitivity lies in the polynomial kernel experiment with a sensitivity value of 99.88%, an accuracy of 99.88%, and a specificity of 99.96%. The method used is able to detect very well and is quite time-effective compared to conventional CNN