Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Sentiment Analysis of the TPKS Law on Twitter: A Comparative Study of Classification Algorithm Performance
No full textThe enactment of Law Number 12 of 2022 concerning the Crime of Sexual Violence (UU TPKS) has sparked significant public discourse on social media, especially on Twitter. This study aims to identify the most effective classification algorithm for analyzing public sentiment regarding the UU TPKS. A total of 2,351 Indonesian-language tweets were collected, preprocessed, and manually labeled into positive and negative sentiments. The Term Frequency–Inverse Document Frequency (TF-IDF) method was used for feature extraction, followed by classification using six algorithms: Naive Bayes (NB), K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and XGBoost. The evaluation results show that SVM and Random Forest achieved the highest accuracy of 85.35%, precision of 0.85, recall of 0.85, and F1-score of 0.83, outperforming other models in handling high-dimensional and imbalanced data. These results demonstrate that the combination of TF-IDF with SVM and Random Forest provides an effective and reliable approach for sentiment analysis of Indonesian-language social media data, particularly in evaluating public responses to socio-legal policies such as the UU TPKS
Household Clustering in West Java Based on Stunting Risk Factors Using K-Modes and K-Prototypes Algorithms
No full textStunting remains one of Indonesia’s most persistent public health challenges, with West Java contributing the highest number of cases due to its large population and regional disparities in household welfare. Identifying household groups vulnerable to stunting is essential for designing targeted interventions that integrate nutrition, sanitation, and socio-economic development. This study introduces a data-driven clustering framework using the K-Modes and K-Prototypes algorithms to classify 22,161 households in West Java based on 26 indicators from the March 2024 National Socioeconomic Survey (SUSENAS), encompassing food security, sanitation, drinking water access, economic conditions, social assistance, and demographics. The K-Modes algorithm was applied to categorical data, while K-Prototypes integrated numerical and categorical variables, with parameter optimization performed using a grid search and the Elbow method. Clustering performance was evaluated through the Silhouette Score, Calinski–Harabasz Index, and Davies–Bouldin Index, followed by a bootstrapped stability analysis employing the Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI). Results show that K-Prototypes outperformed K-Modes, yielding a higher Silhouette Score (0.6681 compared to 0.2922), higher CH Index (13,890.6 compared to 3,976.1), and lower DBI (0.4607 compared to 1.5274), indicating superior compactness and separation. Stability testing confirmed strong robustness, with mean ARI = 0.959 and mean NMI = 0.932 across 50 bootstrap replications. The optimal five-cluster structure identified distinct socioeconomic groups, with the highest stunting risk found among households with low income, limited housing space, inadequate sanitation, and more children under five. The findings highlight the effectiveness of K-Prototypes in modeling mixed-type data and support the design of evidence-based, regionally adaptive stunting reduction strategies aligned with Presidential Regulation No. 72/2021 on the Acceleration of Stunting Reduction
L2IC and MobileViT-XXS for BISINDO Alphabet Recognition
No full textThis study proposes a Landmark-to-Image Conversion (L2IC) approach integrated with the MobileViT-XXS architecture for Indonesian Sign Language (BISINDO) alphabet recognition. The method converts 42 hand keypoints, extracted using MediaPipe Hands into normalized 224×224 grayscale images to capture spatial hand patterns more effectively. These L2IC representations are then used as input to the MobileViT-XXS model, trained for 30 epochs with a learning rate of 0.001. Experimental results show that the model achieves an accuracy and Macro F1-Score of 97.98%, outperforming baseline approaches using raw RGB images and MLP-based classification on numerical keypoints. While the model demonstrates strong performance in controlled offline experiments, further evaluation is required to assess its robustness under real-world dynamic BISINDO usage and deployment on resource-limited devices. These findings indicate that the L2IC representation effectively captures essential spatial information, contributing to high recognition accuracy in static BISINDO hand gesture classification.This study proposes a Landmark-to-Image Conversion (L2IC) approach integrated with the MobileViT-XXS architecture for Indonesian Sign Language (BISINDO) alphabet recognition. The method converts 42 hand keypoints, extracted using MediaPipe Hands into normalized 224×224 grayscale images to capture spatial hand patterns more effectively. These L2IC representations are then used as input to the MobileViT-XXS model, trained for 30 epochs with a learning rate of 0.001. Experimental results show that the model achieves an accuracy and Macro F1-Score of 97.98%, outperforming baseline approaches using raw RGB images and MLP-based classification on numerical keypoints. While the model demonstrates strong performance in controlled offline experiments, further evaluation is required to assess its robustness under real-world dynamic BISINDO usage and deployment on resource-limited devices. These findings indicate that the L2IC representation effectively captures essential spatial information, contributing to high recognition accuracy in static BISINDO hand gesture classification
Implementation of Collaborative Filtering in the Salted Fish Recommendation Process
No full textThe development of e-commerce in the current era has been so rapid that buying and selling transactions are carried out online through various media, including websites and applications. With so many products available in the application, users often feel confused when choosing the product they want to buy, so it takes a long time to choose a product to avoid regret after purchasing it. In this study, a web-based recommendation system was created for the process of recommending salted fish with the aim of making it easier for customers to choose the type of salted fish. The Collaborative Filtering method was used, employing Pearson Correlation as a tool to calculate the similarity value between users, then using Weighted Sum to calculate the prediction value. Collaborative Filtering often experiences the cold start problem, where the system has difficulty providing recommendations to users who do not yet have a transaction history. Therefore, the author proposes a popularity-based strategy as a measure to overcome this problem. Based on testing, the author obtained results of MAE = 0.63 and RMSE = 0.81 based on train-test split results with a data distribution of 80:20, 80% of the dataset for training and 20% of the dataset for testing with an accuracy of 70-80%, indicating that this system works well. This system has been tested using the Blackbox method.The development of e-commerce in the current era has been so rapid that buying and selling transactions are carried out online through various media, including websites and applications. With so many products available in the application, users often feel confused when choosing the product they want to buy, so it takes a long time to choose a product to avoid regret after purchasing it. In this study, a web-based recommendation system was created for the process of recommending salted fish with the aim of making it easier for customers to choose the type of salted fish. The Collaborative Filtering method was used, employing Pearson Correlation as a tool to calculate the similarity value between users, then using Weighted Sum to calculate the prediction value. Collaborative Filtering often experiences the cold start problem, where the system has difficulty providing recommendations to users who do not yet have a transaction history. Therefore, the author proposes a popularity-based strategy as a measure to overcome this problem. Based on testing, the author obtained results of MAE = 0.63 and RMSE = 0.81 based on train-test split results with a data distribution of 80:20, 80% of the dataset for training and 20% of the dataset for testing with an accuracy of 70-80%, indicating that this system works well. This system has been tested using the Blackbox method
Evaluation of the Accuracy and Efficiency of Deep CNN Architecture in Feature Extraction for Guava Disease Classification
No full textThis study analyzes and compares several Deep Convolutional Neural Network (DCNN) architectures to evaluate the balance between classification accuracy and computational efficiency in guava fruit disease detection. A hybrid DCNN–Machine Learning (ML) approach was applied to 3,784 images from the Guava Fruit Disease Dataset using a 10-fold cross-validation scheme and undersampling techniques to address data imbalance. Six DCNN architectures were systematically tested, and the combination of ResNet50 with Artificial Neural Network (ANN) showed the best performance with an accuracy of 0.9979 and an F1-score of 0.9975, surpassing the InceptionV3 baseline (0.9974). In addition to being the most accurate, ResNet50 was also 2.5 times faster in feature extraction than DenseNet201, demonstrating an optimal balance between accuracy and time efficiency. These findings emphasize the importance of analyzing the accuracy-efficiency trade-off in selecting a DCNN architecture and open up opportunities for developing more efficient models for future agricultural image classification applications
Development of an IoT-Based Electric Safety Buoy with Autonomous Navigation System for Coastal Water Rescue Operations
No full textThis research aimed to develop and evaluate an IoT-based electric safety buoy equipped with an autonomous navigation system to support Search and Rescue (SAR) operations in coastal environments. The system integrates dual‐thruster propulsion, GPS and Inertial Measurement Unit (IMU) sensors, IoT telemetry, and a Return-to-Home (RTH) mechanism, enabling both manual and autonomous operation modes. Prototype testing was conducted in a controlled aquatic environment under light wave conditions (10–25 cm) and mild surface currents (0.18–0.32 m/s), with calm weather and unobstructed line-of-sight communication. The buoy was evaluated in both unloaded and 2 kg payload conditions, traveling at an average speed of 1.25–1.35 m/s across test sessions lasting 12–18 minutes. Three predefined GPS waypoints were used to assess navigation accuracy, motion stability, RTH reliability, and telemetry performance. Results show that the autonomous mode achieved a mean positioning error of 1.12 m, a cross-track deviation of 0.35 m, and a waypoint success rate of 96%, outperforming manual navigation by 52%. The RTH function maintained a success rate of 100% under low-battery conditions and 92% during communication loss, while IoT telemetry remained stable up to 200 meters with less than 1% packet loss. These findings confirm that integrating IoT-based telemetry with adaptive autonomous navigation enhances rescue mission efficiency and operational safety, while indicating the need for further validation under more challenging open-sea conditions.This research aimed to develop and evaluate an IoT-based electric safety buoy equipped with an autonomous navigation system to support Search and Rescue (SAR) operations in coastal environments. The system integrates dual‐thruster propulsion, GPS and Inertial Measurement Unit (IMU) sensors, IoT telemetry, and a Return-to-Home (RTH) mechanism, enabling both manual and autonomous operation modes. Prototype testing was conducted in a controlled aquatic environment under light wave conditions (10–25 cm) and mild surface currents (0.18–0.32 m/s), with calm weather and unobstructed line-of-sight communication. The buoy was evaluated in both unloaded and 2 kg payload conditions, traveling at an average speed of 1.25–1.35 m/s across test sessions lasting 12–18 minutes. Three predefined GPS waypoints were used to assess navigation accuracy, motion stability, RTH reliability, and telemetry performance. Results show that the autonomous mode achieved a mean positioning error of 1.12 m, a cross-track deviation of 0.35 m, and a waypoint success rate of 96%, outperforming manual navigation by 52%. The RTH function maintained a success rate of 100% under low-battery conditions and 92% during communication loss, while IoT telemetry remained stable up to 200 meters with less than 1% packet loss. These findings confirm that integrating IoT-based telemetry with adaptive autonomous navigation enhances rescue mission efficiency and operational safety, while indicating the need for further validation under more challenging open-sea conditions
Mapping the Polarity of Tourist Opinions on Indonesian Destinations through Google Maps Reviews Using Supervised Learning Methods
No full textThe advancement of information technology has transformed how individuals seek information and plan their travels, notably through online reviews of tourist attractions on platforms like Google Maps. However, these reviews do not always align with visitors\u27 expectations, necessitating further analysis to comprehend the underlying sentiments. The objective of this research is to inspect the performance of multiple machine learning algorithms in executing sentiment analysis on user generated reviews related to tourist attractions in Indonesia. The algorithms examined include Multinomial Naïve Bayes, Random Forest Classifier, Logistic Regression, Support Vector Machine, K-Nearest Neighbors, and Extra Trees Classifier. The research process encompasses data collection and labeling, data preprocessing, exploratory data analysis (EDA), Word Cloud visualization, feature extraction, classification implementation, and performance evaluation. Experimental results indicate that the K-Nearest Neighbors (KNN) algorithm attain the most accuracy and F1-score of 97%, indicating its effectiveness in categorizing text-based sentiment reviews sourced from the Google Maps platform.The advancement of information technology has transformed how individuals seek information and plan their travels, notably through online reviews of tourist attractions on platforms like Google Maps. However, these reviews do not always align with visitors\u27 expectations, necessitating further analysis to comprehend the underlying sentiments. The objective of this research is to inspect the performance of multiple machine learning algorithms in executing sentiment analysis on user generated reviews related to tourist attractions in Indonesia. The algorithms examined include Multinomial Naïve Bayes, Random Forest Classifier, Logistic Regression, Support Vector Machine, K-Nearest Neighbors, and Extra Trees Classifier. The research process encompasses data collection and labeling, data preprocessing, exploratory data analysis (EDA), Word Cloud visualization, feature extraction, classification implementation, and performance evaluation. Experimental results indicate that the K-Nearest Neighbors (KNN) algorithm attain the most accuracy and F1-score of 97%, indicating its effectiveness in categorizing text-based sentiment reviews sourced from the Google Maps platform
IoT-Based Adaptive Room Temperature Monitoring and Energy Optimization System Using NodeMCU ESP8266
No full textThis study presents the development of an IoT-based room temperature monitoring and AC control system at the Faculty of Engineering, Universitas Malikussaleh, using NodeMCU ESP8266, DHT11 sensor, PIR sensor, and IR LED for real-time automation via a Firebase web interface. The system automatically adjusts AC operation based on room temperature and occupancy, with daily logic resets to accommodate dynamic conditions. Testing conducted over one week demonstrated effective temperature stabilization within 25–26°C with ±2°C fluctuations and significant energy savings by deactivating the AC when the temperature drops below 25°C or the room is unoccupied. The PIR sensor supports a detection range of up to 7 meters, allowing scalability for different room sizes. User evaluation involving five respondents reported satisfaction scores of 4.2 for comfort and energy efficiency, though aspects such as the web interface (3.6) and system information display (2.6) require improvement. Overall, the system effectively enhances energy efficiency, ensures room comfort, and provides flexible control for users, supporting the smart classroom concept. Future development is directed toward the use of more accurate sensors like DHT22 or DS18B20, improved network stability, and integration with virtual assistants for voice-controlled operation.This study presents the development of an IoT-based room temperature monitoring and AC control system at the Faculty of Engineering, Universitas Malikussaleh, using NodeMCU ESP8266, DHT11 sensor, PIR sensor, and IR LED for real-time automation via a Firebase web interface. The system automatically adjusts AC operation based on room temperature and occupancy, with daily logic resets to accommodate dynamic conditions. Testing conducted over one week demonstrated effective temperature stabilization within 25–26°C with ±2°C fluctuations and significant energy savings by deactivating the AC when the temperature drops below 25°C or the room is unoccupied. The PIR sensor supports a detection range of up to 7 meters, allowing scalability for different room sizes. User evaluation involving five respondents reported satisfaction scores of 4.2 for comfort and energy efficiency, though aspects such as the web interface (3.6) and system information display (2.6) require improvement. Overall, the system effectively enhances energy efficiency, ensures room comfort, and provides flexible control for users, supporting the smart classroom concept. Future development is directed toward the use of more accurate sensors like DHT22 or DS18B20, improved network stability, and integration with virtual assistants for voice-controlled operation
Python-Based Linear Regression Modeling of Liquidity and Profitability Ratios as Determinants of Firm Value in Commercial Banks in Indonesia
No full textThe decline in Indonesian banking stocks in early 2024 was caused by foreign capital outflows. A decrease in foreign investment can reduce banks\u27 ability to maintain their capital structure and intensify liquidity pressures. The primary measure for assessing company value is firm value, which can attract investors and indicate potential returns for shareholders. This study aims to analyze the influence of liquidity and profitability on firm value. The sample consists of 55 firm-year observations from the banking sector listed on the Indonesia Stock Exchange (IDX) ranging from 2019 to 2023, obtained using a purposive sampling technique. Data processed with Python using Multiple Linear Regression shows that liquidity has a significant and positive effect on firm value. Similarly, profitability also has a positive and significant effect on firm value. The results of the F-test indicate that liquidity and profitability have a simultaneous influence on firm value. The model demonstrates an outstanding prediction rate, with an R-squared value of 99.8%. The model These findings contribute valuable insights for stakeholders and investors, suggesting that by assessing the strength of liquidity and profitability in financial statements, they can enhance their profit potential.The decline in Indonesian banking stocks in early 2024 was caused by foreign capital outflows. A decrease in foreign investment can reduce banks\u27 ability to maintain their capital structure and intensify liquidity pressures. The primary measure for assessing company value is firm value, which can attract investors and indicate potential returns for shareholders. This study aims to analyze the influence of liquidity and profitability on firm value. The sample consists of 55 firm-year observations from the banking sector listed on the Indonesia Stock Exchange (IDX) ranging from 2019 to 2023, obtained using a purposive sampling technique. Data processed with Python using Multiple Linear Regression shows that liquidity has a significant and positive effect on firm value. Similarly, profitability also has a positive and significant effect on firm value. The results of the F-test indicate that liquidity and profitability have a simultaneous influence on firm value. The model demonstrates an outstanding prediction rate, with an R-squared value of 99.8%. The model These findings contribute valuable insights for stakeholders and investors, suggesting that by assessing the strength of liquidity and profitability in financial statements, they can enhance their profit potential
Pap Smear Image Classification for Cervical Cancer Prediction with Transfer Learning on ResNet101 Architecture
No full textEarly detection of cervical cancer remains a pivotal strategy to improve clinical outcomes and mitigate mortality associated with this disease. This study introduces a robust deep learning framework employing the ResNet101 architecture to facilitate the automated classification of cervical cell images derived from Pap smear examinations. By leveraging transfer learning, the pre-trained ResNet101 model was fine-tuned to extract salient morphological features critical for distinguishing among diverse cervical cell categories. A comprehensive dataset of labeled Pap smear images, systematically expanded through augmentation techniques, was utilized to enhance model generalizability. The proposed approach achieved a remarkable classification accuracy of 99.6%, highlighting its effectiveness in reliably differentiating between normal and abnormal cellular structures. These findings substantiate the promise of deep residual networks coupled with transfer learning as a powerful tool in advancing computer-aided diagnostic systems, thereby reinforcing early screening initiatives for cervical cancer.Early detection of cervical cancer remains a pivotal strategy to improve clinical outcomes and mitigate mortality associated with this disease. This study introduces a robust deep learning framework employing the ResNet101 architecture to facilitate the automated classification of cervical cell images derived from Pap smear examinations. By leveraging transfer learning, the pre-trained ResNet101 model was fine-tuned to extract salient morphological features critical for distinguishing among diverse cervical cell categories. A comprehensive dataset of labeled Pap smear images, systematically expanded through augmentation techniques, was utilized to enhance model generalizability. The proposed approach achieved a remarkable classification accuracy of 99.6%, highlighting its effectiveness in reliably differentiating between normal and abnormal cellular structures. These findings substantiate the promise of deep residual networks coupled with transfer learning as a powerful tool in advancing computer-aided diagnostic systems, thereby reinforcing early screening initiatives for cervical cancer