Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Real-Time Waste Detection System Using YOLOv12 with Transfer Learning
No full textWaste sorting at the source remains a major challenge in Indonesia due to limited public awareness and the absence of accessible tools for waste classification. While YOLO-based object detection has been widely applied for waste detection, the adoption of the latest YOLO architecture in web-based, real-time public-oriented systems remains limited. This study aims to develop and experimentally evaluate a web-based waste detection system using YOLOv12 with a transfer learning approach to classify waste into organic, inorganic, and hazardous (B3) categories along with their subcategories. The system was developed using the Flask framework and supports image upload and real-time camera-based detection. A real-world dataset was annotated and divided into training, validation, and testing sets for experimental evaluation. The proposed model achieved a precision of 0.86, recall of 0.74, [email protected] of 0.83, and [email protected]:0.95 of 0.68, with an average inference time of 0.0187 seconds per image (53.40 FPS). Overall, these results indicate that YOLOv12 with transfer learning provides an effective balance between accuracy and inference speed for web-based real-time waste detection systems, supporting its applicability for practical waste sorting solutions
Numerical Investigation of Nonlinear Parabolic Dynamical Wave Equations Using Modified Variational Iteration Algorithm-II
No full textIn this study, the Modified Variational Iteration Algorithm-II (MVIA-II) is implemented as a robust numerical scheme for solving nonlinear Parabolic partial differential equations. The study focuses on the implementation of an auxiliary parameter h into the correction functional to control the convergence region of the approximate series solution. To validate the efficiency of this semi-numerical approach, two fundamental models arising in mathematical physics and biology are investigated: The Allen-Cahn equation and the Newell-Whitehead equation. The results are compared with exact analytical solutions and other existing numerical methods. The error analysis demonstrates that the proposed algorithm yields high accuracy with minimal computational overhead, making it a promising tool for simulating nonlinear dynamical wave phenomena
Opinion Mining of Pedometer Application Reviews on Google Play Store Using Fine-Tuned IndoBERT-Base
No full textUser reviews on the Google Play Store provide valuable insights into user satisfaction and application performance. However, manual analysis of these reviews is inefficient due to large data volume and the informal characteristics of the Indonesian language. This study proposes an opinion mining approach using a fine-tuned IndoBERT-Base model to classify user sentiments into three classes: positive, neutral, and negative. A total of 1,665 reviews of a Pedometer application were collected, with 1,636 reviews retained after preprocessing. The dataset was divided into training, validation, and test sets using stratified sampling to preserve class distribution. Experimental results show that the proposed model achieves an accuracy of 94.51% and a weighted F1-score of 0.93 on the test set. Despite strong overall performance, the results indicate that class imbalance significantly affects the classification of neutral and negative sentiments. Error analysis reveals that ambiguous expressions and limited samples in minority classes remain challenging for the model. This study demonstrates that fine-tuned IndoBERT-Base is effective for sentiment analysis of Indonesian mobile application reviews while highlighting the importance of addressing imbalanced data in opinion mining tasks.User reviews on the Google Play Store provide valuable insights into user satisfaction and application performance. However, manual analysis of these reviews is inefficient due to large data volume and the informal characteristics of the Indonesian language. This study proposes an opinion mining approach using a fine-tuned IndoBERT-Base model to classify user sentiments into three classes: positive, neutral, and negative. A total of 1,665 reviews of a Pedometer application were collected, with 1,636 reviews retained after preprocessing. The dataset was divided into training, validation, and test sets using stratified sampling to preserve class distribution. Experimental results show that the proposed model achieves an accuracy of 94.51% and a weighted F1-score of 0.93 on the test set. Despite strong overall performance, the results indicate that class imbalance significantly affects the classification of neutral and negative sentiments. Error analysis reveals that ambiguous expressions and limited samples in minority classes remain challenging for the model. This study demonstrates that fine-tuned IndoBERT-Base is effective for sentiment analysis of Indonesian mobile application reviews while highlighting the importance of addressing imbalanced data in opinion mining tasks
Implementasi Video Company Profile PT. Sanindo Multi Tekno Dengan Gabungan Cinematic Dan Motion Graphics Menggunakan Metode MDLC
No full textThis study aims to design a company profile video for PT. Sanindo Multi Tekno using the MDLC method with a combination of cinematic and motion graphic techniques. Development was carried out through six stages: Concept, design, material collection, assembly, testing, and distribution. Production involved interviews, storyboarding, material collection, and editing using CapCut Pro and Adobe After Effects. Evaluation was carried out through alpha testing by multimedia experts and validation from the company. Beta testing involved 35 respondents using the AIDA model and brand awareness analysis. The video was distributed through YouTube and Instagram to expand reach and increase the company\u27s brand awareness.Penelitian ini bertujuan untuk merancang video company profile PT. Sanindo Multi Tekno menggunakan metode MDLC dengan penggabungan teknik cinematic dan motion graphic. Pengembangan dilakukan melalui enam tahap: Concept, design, material collecting, assembly, testing, dan distribution. Produksi melibatkan wawancara, penyusunan storyboard, pengumpulan materi, serta proses editing menggunakan CapCut Pro dan Adobe After Effect. Evaluasi dilakukan melalui pengujian alpha oleh ahli multimedia dan validasi dari pihak perusahaan. Pengujian beta melibatkan 35 responden menggunakan model AIDA serta analisis brand awareness. Video didistribusikan melalui YouTube dan Instagram untuk memperluas jangkauan dan meningkatkan brand awareness perusahaan
Perancangan Game Visual Novel Sebagai Media Pembelajaran Alternatif Pengenalan Asal Usul Aksara Jawa Pada Gen Z
No full textThe visual novel "Aji Saka and the Origin of Javanese Script" is designed as a digital-based interactive media to introduce local culture, especially the origin of Javanese script, to the younger generation, especially Generation Z. Generation Z is a group that was born and grew up in the digital era with characteristics that are very familiar with technology, social media, and interactive applications, so they have a more effective learning style through digital and visual media. This game was developed using the Multimedia Development Life Cycle (MDLC) method which consists of six systematic stages, from concept to distribution. The final product is an educational interactive media that presents the legend of Aji Saka in the form of an interesting visual and textual narrative. User evaluation shows a significant increase in understanding with an average of 4.538/5 (90%), as well as a positive user experience in terms of usability
IoT-Based Water Quality Monitoring and Control System for Koi Fish Ponds
No full textKoi fish (Cyprinus rubrofuscus) require stable water quality to support their health and growth, yet conventional pond water management is generally performed manually and tends to be inefficient and inconsistent. This study aims to design and implement an Internet of Things (IoT)-based water quality monitoring and control system for koi fish ponds. The proposed system integrates an ESP32 microcontroller with pH, turbidity, ultrasonic, and water level sensors to monitor pond conditions in real time and support controlled water drainage and refilling through a web-based interface. Sensor data are transmitted to Firebase Cloud, enabling remote monitoring and control via an internet connection. System testing was conducted on four koi ponds with ten measurements for each parameter, resulting in forty data samples per parameter. The experimental results show that the sensors provide stable measurements with average error values below 3%, and the system demonstrates a response time of approximately 1–2 seconds under stable network conditions. These results indicate that the developed system is capable of supporting effective water quality monitoring and control while reducing reliance on continuous manual supervision in koi pond management.Koi fish (Cyprinus rubrofuscus) require stable water quality to support their health and growth, yet conventional pond water management is generally performed manually and tends to be inefficient and inconsistent. This study aims to design and implement an Internet of Things (IoT)-based water quality monitoring and control system for koi fish ponds. The proposed system integrates an ESP32 microcontroller with pH, turbidity, ultrasonic, and water level sensors to monitor pond conditions in real time and support controlled water drainage and refilling through a web-based interface. Sensor data are transmitted to Firebase Cloud, enabling remote monitoring and control via an internet connection. System testing was conducted on four koi ponds with ten measurements for each parameter, resulting in forty data samples per parameter. The experimental results show that the sensors provide stable measurements with average error values below 3%, and the system demonstrates a response time of approximately 1–2 seconds under stable network conditions. These results indicate that the developed system is capable of supporting effective water quality monitoring and control while reducing reliance on continuous manual supervision in koi pond management
Application of SARIMA, GRU, and Prophet for Capturing Seasonal Patterns in Consumer Price Inflation
No full textSeasonal dynamics make inflation forecasting challenging in emerging economies where holiday effects, regulated prices, and supply shocks interact. This study models Indonesia’s monthly consumer price inflation (CPI) using official data from Statistics Indonesia (May 2006–April 2025) and evaluates three forecasting paradigms: a classical seasonal baseline (SARIMA), a decomposable model with trend–seasonality components (Prophet), and a neural sequence learner (GRU). A 10-fold sliding window design is employed to preserve temporal order. Performance is assessed with RMSE, MAE, and MASE, summarized across folds with boxplots and statistical descriptives (means, standard deviations, and 95% confidence intervals). Across folds and metrics, Prophet consistently achieves the lowest error and the tightest dispersion, GRU ranks second with competitive accuracy and stable variance, and SARIMA remains a transparent yet weaker benchmark. MASE values below one for Prophet (and generally for GRU) indicate improvements over a naïve baseline. Practically, Prophet’s decompositions support policy communication by linking forecast movements to interpretable components (e.g., Ramadan/Eid and year-end effects), while GRU is useful during more nonlinear or volatile periods; SARIMA remains valuable for diagnostics in stable regimes
Outperforming DNN Using MLP in Water Quality Assessment for Aquaculture
No full textAquaculture production relies heavily on stable water quality conditions, requiring accurate and efficient assessment methods to support early environmental monitoring and sustainable management. Although deep neural network models have been widely applied to water quality classification, their high computational complexity often limits their applicability in real-time and resource-constrained aquaculture systems. This study aims to evaluate whether a systematically optimized Multilayer Perceptron can outperform a reported deep neural network benchmark in aquaculture water quality assessment while maintaining computational efficiency. The study adopts a structured methodology involving dataset characterization, extreme outlier removal, feature normalization, and stratified data partitioning. A single-hidden-layer Multilayer Perceptron is trained using a feedforward backpropagation learning process, with systematic exploration of hidden neuron configurations and training epochs to identify the optimal architecture. Model performance is evaluated using multiple classification metrics, including accuracy, precision, recall, F1-score, confusion matrix analysis, and receiver operating characteristic and precision–recall curves. Results indicate that the optimal Multilayer Perceptron configuration, consisting of 80 hidden neurons and 200 training epochs, achieves an accuracy of 96.62%, surpassing the deep neural network benchmark accuracy of 95.69%. The proposed model demonstrates strong class-level performance, clear separation between water quality categories, stable convergence behavior, and reduced computational overhead compared to deeper architectures. These findings highlight that increasing model depth does not necessarily improve predictive performance for heterogeneous aquaculture datasets. In conclusion, this study provides empirical evidence that a well-optimized shallow neural network can outperform deeper models in aquaculture water quality assessment. The results emphasize the importance of model parsimony and systematic hyperparameter optimization, offering a practical and efficient solution for real-time aquaculture water quality monitoring applications
Calibration and Applied Statistical Modeling Using Logistic Regression on the UCI Heart Disease Dataset
No full textAccurate and well-calibrated heart disease risk prediction is essential for supporting medical decision-making. This study analyzes Logistic Regression as an applied statistical model for heart disease prediction using the UCI Heart Disease dataset. Beyond discrimination metrics, we explicitly focus on probability reliability by evaluating calibration through the Brier score, calibration slope, and intercept, and by quantifying the impact of post-hoc calibration (isotonic regression and Platt scaling) on both calibration and discrimination. Model validation was conducted using stratified 5-fold cross-validation with AUROC, AUPRC, accuracy, and F1-score as evaluation metrics. The results show that Logistic Regression achieved competitive performance (AUROC 0.903; AUPRC 0.911; Accuracy 0.822; F1-score 0.835) with well-calibrated probability estimates relative to Random Forest and Gradient Boosting under the evaluated setting. Feature importance analysis using permutation methods identified chest pain type, number of major vessels (ca), ST depression (oldpeak), and exercise-induced angina (exang) as key predictors consistent with clinical literature. These findings indicate that simple applied statistical modeling, when paired with rigorous calibration assessment, can provide interpretable risk estimates that are more suitable for threshold-based decision support in early heart disease screening
A Fine-Tuned Transfer Learning Vision Transformer Framework for Lungs X-Ray Image Classification
No full textLung diseases constitute a significant source of morbidity and therefore require diagnostic frameworks that provide both high accuracy and operational efficiency. This study proposes the development of a Vision Transformer (ViT)-based classification model for lung X-ray images, employing transfer learning and fine-tuning techniques to improve detection performance across five disease categories. Experimental results demonstrate stable and effective model convergence, as reflected by the consistent decrease in loss metrics throughout the learning process. Evaluation on an independent test dataset shows that the proposed approach achieves an accuracy of 0.958, indicating strong and balanced generalization performance. Further analysis using a confusion matrix reveals that the ViT model is capable of recognizing subtle and complex radiographic patterns with low misclassification rates, particularly achieving high recall for major pathological classes, which is critical for minimizing false negatives in clinical screening scenarios. Overall, this study demonstrates that the application of transfer learning with fine-tuning on a Vision Transformer architecture yields competitive performance for multi-class lung X-ray classification when trained on a balanced dataset. These findings are consistent with prior evidence highlighting the effectiveness of ViT in capturing global contextual information in medical imaging tasks