Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Topic Clustering of Student Complaints Based on Semantic Meaning Using the indoBERT and K-Means Models
Get PDFThis study applies Natural Language Processing (NLP) technology to extract and cluster information from student complaint text data. The model used is IndoBERT, a variant of BERT (Bidirectional Encoder Representations from Transformers) that has been adapted for the Indonesian language. The main objective of this research is to perform topic clustering based on semantic similarity. The process begins with data collection and cleaning, followed by tokenization and text normalization. Each complaint is transformed into a vector representation through IndoBERT embeddings, which are then used as input for the K-Means clustering algorithm. Evaluation is conducted using various metrics, and the results of the Silhouette Score and Elbow Method indicate that the optimal number of clusters is four. Cluster visualization using the t-distributed Stochastic Neighbor Embedding (t-SNE) method reinforces these findings by displaying four fairly distinct groups of complaints, although one cluster appears dispersed and less well-defined, indicating possible topic overlap. The quality of topics within each cluster is evaluated using the Topic Coherence (c_v) metric, where Cluster 3 achieved the highest score of 0.7084. The topics in this cluster highlight critical issues such as campus facilities, lecturer quality, and information delivery systems. Overall, the four resulting clusters reflect central themes: Facilities, Expectations or Impressions, Services, and Academic Lectures. These results are expected to serve as a reference for institutions in formulating service improvement policies based on student complaint analysis.This study applies Natural Language Processing (NLP) technology to extract and cluster information from student complaint text data. The model used is IndoBERT, a variant of BERT (Bidirectional Encoder Representations from Transformers) that has been adapted for the Indonesian language. The main objective of this research is to perform topic clustering based on semantic similarity. The process begins with data collection and cleaning, followed by tokenization and text normalization. Each complaint is transformed into a vector representation through IndoBERT embeddings, which are then used as input for the K-Means clustering algorithm. Evaluation is conducted using various metrics, and the results of the Silhouette Score and Elbow Method indicate that the optimal number of clusters is four. Cluster visualization using the t-distributed Stochastic Neighbor Embedding (t-SNE) method reinforces these findings by displaying four fairly distinct groups of complaints, although one cluster appears dispersed and less well-defined, indicating possible topic overlap. The quality of topics within each cluster is evaluated using the Topic Coherence (c_v) metric, where Cluster 3 achieved the highest score of 0.7084. The topics in this cluster highlight critical issues such as campus facilities, lecturer quality, and information delivery systems. Overall, the four resulting clusters reflect central themes: Facilities, Expectations or Impressions, Services, and Academic Lectures. These results are expected to serve as a reference for institutions in formulating service improvement policies based on student complaint analysis
Sentiment Classification of Indonesian-Language Roblox Reviews Using IndoBERT with SMOTE Optimization
Get PDFRoblox is a community-based gaming platform that is extremely popular among users of various age groups. Millions of user reviews available on the platform contain valuable information regarding user satisfaction, expectations, and criticisms of the gameplay experience. To extract insights from these reviews, a reliable natural language processing (NLP) approach tailored to the local language context is essential. This study aims to classify sentiments in Indonesian-language user reviews of Roblox into three categories: positive, negative, and neutral. The model used is IndoBERT, a transformer-based model specifically trained to understand the structure and vocabulary of the Indonesian language. One of the main challenges in this study is the imbalance in the number of data points across sentiment classes. To address this, the SMOTE (Synthetic Minority Over-sampling Technique) method is applied to strengthen the representation of minority classes. The dataset consists of thousands of reviews that have been manually labeled by annotators. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics. The results show that the combination of IndoBERT and SMOTE provides significant improvements compared to the baseline approach without oversampling. This research contributes to the development of automated sentiment analysis systems in the Indonesian language, which can be applied across various digital platforms. The implementation of this model can assist game developers and product analysts in efficiently understanding user opinions, thereby driving improvements in service quality and user satisfaction in the future.Roblox is a community-based gaming platform that is extremely popular among users of various age groups. Millions of user reviews available on the platform contain valuable information regarding user satisfaction, expectations, and criticisms of the gameplay experience. To extract insights from these reviews, a reliable natural language processing (NLP) approach tailored to the local language context is essential. This study aims to classify sentiments in Indonesian-language user reviews of Roblox into three categories: positive, negative, and neutral. The model used is IndoBERT, a transformer-based model specifically trained to understand the structure and vocabulary of the Indonesian language. One of the main challenges in this study is the imbalance in the number of data points across sentiment classes. To address this, the SMOTE (Synthetic Minority Over-sampling Technique) method is applied to strengthen the representation of minority classes. The dataset consists of thousands of reviews that have been manually labeled by annotators. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics. The results show that the combination of IndoBERT and SMOTE provides significant improvements compared to the baseline approach without oversampling. This research contributes to the development of automated sentiment analysis systems in the Indonesian language, which can be applied across various digital platforms. The implementation of this model can assist game developers and product analysts in efficiently understanding user opinions, thereby driving improvements in service quality and user satisfaction in the future
The Influence of Employee Engagement, Psychological Capital, and Work-life Balance on Turnover Intention through the mediation of Job Satisfaction of Private Employees in Jabodetabek
Get PDFThis study aims to analyze the influence of employee engagement, psychological capital, and work-life balance on turnover intention with job satisfaction as a mediating variable. In this study, a quantitative approach was used on 439 private employees in Greater Jakarta. This approach uses the Partial Least Squares–Structural Equation Modelling (PLS-SEM) analysis technique. Research shows that employee engagement, psychological capital, and work-life balance have a significant positive effect on employee job satisfaction. In addition, it is proven that job satisfaction has a significant negative effect on turnover intention. All three independent variables have a significant effect on turnover intention, either directly or indirectly through job satisfaction as a mediator. Employee engagement is reported to be the most significant factor in increasing job satisfaction and reducing the desire to leave work. These results support the theory of Job Demands-Resources and Social Exchange Theory, both of which emphasize the importance of psychological resources and social relationships in work. In addition, this study also found that there were differences in views between generations about the variables discussed. This research provides theoretical contributions and practical benefits to the development of sustainable employee retention strategies
Pengendalian Kualitas Produksi Terhadap Penurunan Produk Cacat Pada Section C&C Post Process di PT. X Menggunakan Six Sigma
No full textKualitas merupakan poin kritis pada wire harness, jika terjadi kelolosan cacat akan berdampak terhadap aliran listrik dan operasional mesin mobil. Permasalahan pada tingkat produk cacat section C&C Post Process Maret 2023 yang melampaui 27,2 PPM dari target yang diharapkan pada periode tersebut. Tujuan dari penelitian ini adalah mengidentifikasi factor yang menyebabkan cacat produk dan mengetahui factor dominan penyebab cacat pada hasil produksi. Hal ini mendasari penelitian ini untuk menganalisis akar permasalahan pada section ini. Penelitian ini menggunakan metode Six Sigma dengan pendekatan DMAIC untuk mengukur dan menganalisis kualitas proses produksi yang berjalan. Analisis permasalahan ini difokuskan terhadap jenis cacat strand out dengan persentase kejadian tertinggi dan permasalahan diuraikan dengan faktor kegagalan 4M pada FTA. Selanjutnya uraian akar penyebab pada FTA dianalisis kembali pada FMEA untuk penentuan penerapan prioritas perbaikan. Prioritas perbaikan penelitian ini dilakukan terhadap 3 penyebab kegagalan dengan RPN tertinggi. Tindakan perbaikan yang diterapkan menunjukkan adanya penurunan terhadap perolehan cacat strand out sebesar 67,5%. Namun, perolehan produk cacat pada April 2023 belum mencapai penurunan yang ditargetkan sebanyak 141 unit atau dengan selisih level sigma 0,21. Hasil ini menunjukkan pentingnya penerapan continuous improvement untuk meminimalisir risiko terjadinya kembali cacat secara berkala dan target tingkat produk cacat dapat tercapai. Rekomendasi perbaikan antara lain pembuatan plotting fixed applicator setiap mesin, pembuatan draft stnadar kerja penggunaan magnifying glass menerangi celah crimper dan mengecek hasil crimper pada wire serta memposisikan clemp lurus dengan rootbar agar insert wire dilakukan dalam posisi lurus
The Analysis of the Effectiveness of SMPIT Tunas Cendekia Profile Videos as Digital Promotion Media: Mixed Method Approach with Epic Model, Qualitative Interviews, and Youtube Analytics
No full textIn the era of digital competition, educational institutions are required to adopt innovative and technology-based promotional strategies. This study aims to evaluate the effectiveness of a school profile video of SMPIT Tunas Cendekia as a digital promotional medium to support student enrollment campaigns. A mixed-method approach was applied by integrating the EPIC Model (Empathy, Persuasion, Impact, Communication), qualitative interviews, and YouTube Analytics. Quantitative data were collected from 66 respondents using a Likert-scale questionnaire based on the EPIC Model, while qualitative insights were gathered through in-depth interviews with prospective parents and school representatives. Additionally, YouTube performance metrics such as view count, watch time, and click-through rate (CTR) were analyzed. The findings show that all four EPIC dimensions scored in the "Highly Effective" category, with an overall EPIC Rate of 4.26. Interview results confirmed strong emotional resonance, increased interest, and a positive perception of the school’s image. YouTube data indicated solid engagement, with a CTR of 4.39% and an average watch duration of 2 minutes and 44 seconds. The integration of audience perception, qualitative insight, and digital metrics provides comprehensive evidence that profile videos serve as an effective communication tool in educational marketing. This study recommends content enhancements based on audience needs to improve future promotional efforts
Forecasting coconut production in West Aceh Using GIS and SARIMAX
No full textCoconut (Cocos nucifera) is a strategic commodity for agro-industrial development in Indonesia, especially in Sumatra, which is home to 34.5% of national coconut plantations. One of the major producers, with a coastal geography and tropical climate that is highly suitable for coconut plantations, Aceh Barat, is currently facing the threat of degradation of coconut plantation land loss due to the government\u27s Regional Action Plan for Sustainable Palm Oil Plantations (RAD KSB Aceh 2023-2026). This study aims to look at the total coconut plantation land by integrating geospatial analysis (QGIS) and SARIMAX time series modelling to map coconut plantations in 2024, estimate production trends, and assess the viability of the agro-industry amidst land use conflicts. Results from mapping with QGIS software showed a drastic decrease in coconut area from 3,330.25 hectares in 2022 to 928.2 hectares in 2024. The reduction in coconut plantation area is signalled by RAD KSB\u27s oil palm expansion target of 1,078,728 hectares by 2026. In addition, the results of the mapping obtained several sub-districts with the largest contribution in West Aceh, namely Kaway XVI (234.82 ha) and Muereubo (217.46 ha) of coconut plantation area, while Bubon (16.67 ha) and West Woyla (38.42 ha) experienced significant land conversion. The study also calculated coconut fruit production of 1,229,267 kg (1,229 tonnes) per month from 12 sub-districts, and generated revenue from selling only coconuts of IDR 2.23 billion. SARIMAX forecasts showed high accuracy (RMSE: 700-704; MAPE: 0.19-1.05%) for 10 sub-districts, except Bubon (MAPE: 2.13%) and West Woyla (MAPE: 1.05%) due to data volatility. Furthermore, projections for the next five periods were carried out and obtained results, namely, Period 1 (104,425.88 kg), Period 2 (94,851.07 kg), Period 3 (97,399.50 kg), Period 4 (96,721.21 kg), and Period 5 (96,901.75 kg) which were dominated by stable production in the core area of Kaway XVI: 311,870 kg/month, but volatile in smaller areas. Spatial analysis prioritises Samatiga (58.53 ha) and Arongan Lambalek (79.27 ha) for agro-industrial development, with potential for value-added products.Coconut (Cocos nucifera) is a strategic commodity for agro-industrial development in Indonesia, especially in Sumatra, which is home to 34.5% of national coconut plantations. One of the major producers, with a coastal geography and tropical climate that is highly suitable for coconut plantations, Aceh Barat, is currently facing the threat of degradation of coconut plantation land loss due to the government\u27s Regional Action Plan for Sustainable Palm Oil Plantations (RAD KSB Aceh 2023-2026). This study aims to look at the total coconut plantation land by integrating geospatial analysis (QGIS) and SARIMAX time series modelling to map coconut plantations in 2024, estimate production trends, and assess the viability of the agro-industry amidst land use conflicts. Results from mapping with QGIS software showed a drastic decrease in coconut area from 3,330.25 hectares in 2022 to 928.2 hectares in 2024. The reduction in coconut plantation area is signalled by RAD KSB\u27s oil palm expansion target of 1,078,728 hectares by 2026. In addition, the results of the mapping obtained several sub-districts with the largest contribution in West Aceh, namely Kaway XVI (234.82 ha) and Muereubo (217.46 ha) of coconut plantation area, while Bubon (16.67 ha) and West Woyla (38.42 ha) experienced significant land conversion. The study also calculated coconut fruit production of 1,229,267 kg (1,229 tonnes) per month from 12 sub-districts, and generated revenue from selling only coconuts of IDR 2.23 billion. SARIMAX forecasts showed high accuracy (RMSE: 700-704; MAPE: 0.19-1.05%) for 10 sub-districts, except Bubon (MAPE: 2.13%) and West Woyla (MAPE: 1.05%) due to data volatility. Furthermore, projections for the next five periods were carried out and obtained results, namely, Period 1 (104,425.88 kg), Period 2 (94,851.07 kg), Period 3 (97,399.50 kg), Period 4 (96,721.21 kg), and Period 5 (96,901.75 kg) which were dominated by stable production in the core area of Kaway XVI: 311,870 kg/month, but volatile in smaller areas. Spatial analysis prioritises Samatiga (58.53 ha) and Arongan Lambalek (79.27 ha) for agro-industrial development, with potential for value-added products
Implementation of Clustering Method Using K-Means Algorithm for Grouping BPJS Health Patient Medical Record Data
No full textClustering medical record data of BPJS Health patients is essential in supporting data-driven decision-making in hospitals. This study aims to implement the K-Means algorithm to cluster patient medical records at RSUD Simeulue based on BPJS class and patient address variables. The data were first normalized using the Z-Score method to standardize variable scales, followed by the iterative application of the K-Means algorithm until convergence was reached at the sixth iteration. The study employed three Cluster, namely Cluster 1 (Very Many), Cluster 2 (Many), and Cluster 3 (Not Many). The final results show that Cluster 1 contains 258 patients from Class 1 and 292 from Class 2; Cluster 2 consists of 296 patients from Class 2; and Cluster 3 includes 101 patients from Class 1, 115 from Class 2, and 148 from Class 3. In addition to classification by BPJS class, clustering based on patient address revealed a dominant distribution from Simeulue Timur, Teluk Dalam, and Teupah Selatan sub-districts. The clustering results were implemented into a web-based information system using the Laravel framework and MySQL database, enabling hospital administrators to visualize and analyze patient data effectively. This study demonstrates that the K-Means algorithm can be effectively applied in classifying medical record data to support healthcare management decision-making.Clustering medical record data of BPJS Health patients is essential in supporting data-driven decision-making in hospitals. This study aims to implement the K-Means algorithm to cluster patient medical records at RSUD Simeulue based on BPJS class and patient address variables. The data were first normalized using the Z-Score method to standardize variable scales, followed by the iterative application of the K-Means algorithm until convergence was reached at the sixth iteration. The study employed three Cluster, namely Cluster 1 (Very Many), Cluster 2 (Many), and Cluster 3 (Not Many). The final results show that Cluster 1 contains 258 patients from Class 1 and 292 from Class 2; Cluster 2 consists of 296 patients from Class 2; and Cluster 3 includes 101 patients from Class 1, 115 from Class 2, and 148 from Class 3. In addition to classification by BPJS class, clustering based on patient address revealed a dominant distribution from Simeulue Timur, Teluk Dalam, and Teupah Selatan sub-districts. The clustering results were implemented into a web-based information system using the Laravel framework and MySQL database, enabling hospital administrators to visualize and analyze patient data effectively. This study demonstrates that the K-Means algorithm can be effectively applied in classifying medical record data to support healthcare management decision-making
Experimental Evaluation of Wazuh-Grafana Integration for Real-Time Cyber Threat Detection in Resource-Constrained Environments
No full textThis research evaluates the performance of integrating Wazuh, an open-source Security Information and Event Management (SIEM) platform, with Grafana, a real-time visualization tool, for cyber threat detection in resource-constrained environments. The objective is to assess detection accuracy, false positive rates, response times, and system efficiency under controlled experimental conditions. The testbed consisted of two virtual private servers (4 vCPUs, 4–8 GB RAM, 38–50 GB storage) and employed the CIC-IDS2017 dataset as a benchmark for simulating three representative attacks: brute-force, malware injection, and webshell exploitation. The results showed that the integrated system achieved 100% detection accuracy with 0% false positives across 30 trials, with an average total detection time of 3033 ms. Resource utilization remained low, with CPU usage below 35% and memory consumption under 25%, confirming feasibility for mid-range servers typical of small institutions. While these results underscore the system’s efficiency, the findings must be interpreted within the limitations of a laboratory environment where predefined signatures were used. Performance in real-world networks with diverse traffic and unknown threats may differ, and further validation is required. This study makes two key contributions: (1) it provides the first structured quantitative benchmark of Wazuh-Grafana integration in constrained environments using a standardized dataset, and (2) it offers practical recommendations for small and medium-sized institutions, including minimum system requirements and guidelines for dashboard configuration. These findings reinforce the role of open-source solutions as affordable, adaptive, and effective alternatives to commercial SIEM systems, particularly for organizations with limited cybersecurity budgets.This research evaluates the performance of integrating Wazuh, an open-source Security Information and Event Management (SIEM) platform, with Grafana, a real-time visualization tool, for cyber threat detection in resource-constrained environments. The objective is to assess detection accuracy, false positive rates, response times, and system efficiency under controlled experimental conditions. The testbed consisted of two virtual private servers (4 vCPUs, 4–8 GB RAM, 38–50 GB storage) and employed the CIC-IDS2017 dataset as a benchmark for simulating three representative attacks: brute-force, malware injection, and webshell exploitation. The results showed that the integrated system achieved 100% detection accuracy with 0% false positives across 30 trials, with an average total detection time of 3033 ms. Resource utilization remained low, with CPU usage below 35% and memory consumption under 25%, confirming feasibility for mid-range servers typical of small institutions. While these results underscore the system’s efficiency, the findings must be interpreted within the limitations of a laboratory environment where predefined signatures were used. Performance in real-world networks with diverse traffic and unknown threats may differ, and further validation is required. This study makes two key contributions: (1) it provides the first structured quantitative benchmark of Wazuh-Grafana integration in constrained environments using a standardized dataset, and (2) it offers practical recommendations for small and medium-sized institutions, including minimum system requirements and guidelines for dashboard configuration. These findings reinforce the role of open-source solutions as affordable, adaptive, and effective alternatives to commercial SIEM systems, particularly for organizations with limited cybersecurity budgets
Image-Based Classification of Healthy and Unhealthy Goats Using ResNet-18 Deep Learning Model
No full textEarly detection of livestock health conditions is a critical factor in maintaining animal productivity and welfare. This study aims to develop an image-based classification system for identifying healthy and unhealthy goats using deep learning techniques. The dataset of goat images was obtained from Roboflow and processed through a series of augmentation, normalization, and feature extraction stages using the ResNet-18 convolutional neural network architecture pretrained on ImageNet. The dataset was divided into training and testing sets with a 70:30 stratified split to ensure balanced class distribution. To address class imbalance, a random undersampling technique was applied. The model was trained using optimally tuned hyperparameters, including the Adam optimizer, cross-entropy loss function, a batch size of 32, and 20 epochs. Evaluation results showed that the model achieved an accuracy of 95.97%, with a precision of 96.22%, recall of 95.97%, and F1-score of 95.92%. The confusion matrix and evaluation curves demonstrated that the model is both stable and reliable. These findings indicate that the proposed system has strong potential to be implemented in automated and real-time livestock health monitoring applications, providing a fast, accurate, and non-invasive solution for precision livestock farming.Early detection of livestock health conditions is a critical factor in maintaining animal productivity and welfare. This study aims to develop an image-based classification system for identifying healthy and unhealthy goats using deep learning techniques. The dataset of goat images was obtained from Roboflow and processed through a series of augmentation, normalization, and feature extraction stages using the ResNet-18 convolutional neural network architecture pretrained on ImageNet. The dataset was divided into training and testing sets with a 70:30 stratified split to ensure balanced class distribution. To address class imbalance, a random undersampling technique was applied. The model was trained using optimally tuned hyperparameters, including the Adam optimizer, cross-entropy loss function, a batch size of 32, and 20 epochs. Evaluation results showed that the model achieved an accuracy of 95.97%, with a precision of 96.22%, recall of 95.97%, and F1-score of 95.92%. The confusion matrix and evaluation curves demonstrated that the model is both stable and reliable. These findings indicate that the proposed system has strong potential to be implemented in automated and real-time livestock health monitoring applications, providing a fast, accurate, and non-invasive solution for precision livestock farming
Myopia Identification by Fundus Photo Image Classification Using Convolutional Neural Network
No full textMyopia is a significant vision problem worldwide, requiring early detection to prevent further damage. This study aims to develop an image classification model using a Convolutional Neural Network (CNN) to identify myopia based on fundus images. The dataset used was 124,749 fundus images, divided into 80% for training and 20% for testing. The applied architecture was EfficientNetB0, chosen for its ability to achieve high performance with efficient computation. Experimental results showed that this model successfully achieved a classification accuracy of 97% in distinguishing between myopic and non-myopic images. These findings demonstrate the potential of CNN, especially EfficientNetB0, as a diagnostic tool for automatic myopia identification, which can accelerate the detection process and improve the accuracy of clinical diagnosis.Myopia is a significant vision problem worldwide, requiring early detection to prevent further damage. This study aims to develop an image classification model using a Convolutional Neural Network (CNN) to identify myopia based on fundus images. The dataset used was 124,749 fundus images, divided into 80% for training and 20% for testing. The applied architecture was EfficientNetB0, chosen for its ability to achieve high performance with efficient computation. Experimental results showed that this model successfully achieved a classification accuracy of 97% in distinguishing between myopic and non-myopic images. These findings demonstrate the potential of CNN, especially EfficientNetB0, as a diagnostic tool for automatic myopia identification, which can accelerate the detection process and improve the accuracy of clinical diagnosis