Jurnal Politeknik Negeri Batam (PoliBatam)
Not a member yet
3001 research outputs found
Sort by
Pemanfaatan Machine Learning untuk Menganalisis Praktik ESG: Studi Kasus pada Bank Berkinerja Unggul di Indonesia
No full textThis research related to the discover ESG practices by Top Performing Banks in Indonesia based on their Return on Asset (ROA) and Net Interest Margin (NIM). Due to the shifting the interest of the investor and the world condtion. Recently, investors have shown a growing inclination to incorporate ESG (Environmental, Social, and Governance) considerations into their decision-making processes. This trend is driven by a heightened awareness of environmental risks, particularly those linked to sustainability. In response to climate change, there is also a growing public interest in adopting more sustainable practices. Due to the broad and sometimes ambiguous definition of ESG, identifying the specific practices that should be implemented by banks remains challenging. Research on frameworks to guide these actions is still limited. For instance, explores the role of knowledge management in facilitating the integration of ESG factors. This research aims to create an ESG knowledge discovery framework with Natural Language Process (NLP) based on DBSCAN results of several companies which have good financial performance such as Return on Asset (ROA) and Net Interest Margin (NIM). The result will be analyzed by NLP, especially TF-IDF (Term Frequency - Inverse Document Frequency) and visualized by Word Cloud. The priority of Top Performing Bank in Indonesia related to the environment indicator conduct the activity which focus on carbon, green, and water. For social indicator, the main focus are employee, training, and financial. Lastly, for governance indicator, they are most prominent activity to the security, data, and policy. Overall, leading banks in Indonesia tend to prioritize environmental aspects in their operations. Companies with strong environmental initiatives have been shown to positively impact their earnings and financial performance
Stroke Risk Classification Using the Ensemble Learning Method of XGBoost and Random Forest
No full textStroke is a leading cause of global death and disability. This study proposes a stroke risk classification model using ensemble learning that combines Random Forest and XGBoost algorithms. A Kaggle dataset with 5110 samples (249 stroke, 4861 non-stroke) presented significant class imbalance. To address this, a comprehensive preprocessing pipeline was implemented, including feature encoding, feature scaling, feature selection using ANOVA F-test, outlier handling with Z-Score and IQR methods, and missing value imputation using MICE. The SMOTE-ENN approach was applied to handle class imbalance, resulting in a more balanced sample distribution. The dataset was split into 80% training and 20% testing data (hold-out test) to ensure objective evaluation. Hyperparameter optimization was performed using Bayesian optimization, while model evaluation employed stratified K-fold cross-validation to prevent overfitting. Validation on the hold-out test set demonstrated exceptional ensemble model performance with an AUC of 0.99, 98% accuracy, 98% precision, and 98% recall. Feature importance analysis identified average glucose level and age as the strongest stroke risk predictors. The proposed approach significantly improved predictive accuracy compared to previous research, demonstrating the effectiveness of ensemble learning and preprocessing methods in developing reliable, high-performing machine learning models for early stroke risk assessment.Stroke is a leading cause of global death and disability. This study proposes a stroke risk classification model using ensemble learning that combines Random Forest and XGBoost algorithms. A Kaggle dataset with 5110 samples (249 stroke, 4861 non-stroke) presented significant class imbalance. To address this, a comprehensive preprocessing pipeline was implemented, including feature encoding, feature scaling, feature selection using ANOVA F-test, outlier handling with Z-Score and IQR methods, and missing value imputation using MICE. The SMOTE-ENN approach was applied to handle class imbalance, resulting in a more balanced sample distribution. The dataset was split into 80% training and 20% testing data (hold-out test) to ensure objective evaluation. Hyperparameter optimization was performed using Bayesian optimization, while model evaluation employed stratified K-fold cross-validation to prevent overfitting. Validation on the hold-out test set demonstrated exceptional ensemble model performance with an AUC of 0.99, 98% accuracy, 98% precision, and 98% recall. Feature importance analysis identified average glucose level and age as the strongest stroke risk predictors. The proposed approach significantly improved predictive accuracy compared to previous research, demonstrating the effectiveness of ensemble learning and preprocessing methods in developing reliable, high-performing machine learning models for early stroke risk assessment
Public Sentiment Analysis on the Boycott Israel Movement on Platform X Using Random Forest and Logistic Regression Algorithms
No full textThis research aims to analyze public sentiment toward the boycott movement against Israel on the X platform by applying Random Forest and Logistic Regression algorithms. The study uses 616 tweets collected through web crawling with relevant keywords such as "Boikot", "Israel", and "Palestine", covering the period from March 1, 2023 to January 30, 2025. The dataset underwent preprocessing including cleaning, normalization, stopword removal, tokenization, and stemming. Sentiment labeling was conducted both manually, categorizing the data into positive, negative, and neutral classes. TF-IDF was used for feature weighting. The data was split into 80% training and 20% testing. The Random Forest model achieved an accuracy of 70%, while Logistic Regression reached 68%. Both models showed higher accuracy in predicting positive sentiment compared to negative and neutral. The results suggest that public opinion on the boycott movement on social media tends to be supportive, with “Boikot,” “Israel,” and “Palestine” being the most dominant terms. Random Forest performed slightly better in classification, though improvements are needed in recognizing non-positive sentiments.This research aims to analyze public sentiment toward the boycott movement against Israel on the X platform by applying Random Forest and Logistic Regression algorithms. The study uses 616 tweets collected through web crawling with relevant keywords such as "Boikot", "Israel", and "Palestine", covering the period from March 1, 2023 to January 30, 2025. The dataset underwent preprocessing including cleaning, normalization, stopword removal, tokenization, and stemming. Sentiment labeling was conducted both manually, categorizing the data into positive, negative, and neutral classes. TF-IDF was used for feature weighting. The data was split into 80% training and 20% testing. The Random Forest model achieved an accuracy of 70%, while Logistic Regression reached 68%. Both models showed higher accuracy in predicting positive sentiment compared to negative and neutral. The results suggest that public opinion on the boycott movement on social media tends to be supportive, with “Boikot,” “Israel,” and “Palestine” being the most dominant terms. Random Forest performed slightly better in classification, though improvements are needed in recognizing non-positive sentiments
Geographic Information System for Mapping Accommodation Locations in Lhokseumawe City Using the AHP Method and Dijkstra\u27s Algorithm
No full textThis study aims to develop a web-based Geographic Information System (GIS) to provide recommendations for the best accommodation and the fastest route to the accommodation location in Lhokseumawe City. The Analytical Hierarchy Process (AHP) method is used to determine the priority of accommodation based on five main criteria, namely price, public facilities, cleanliness, security, and year founded. The Dijkstra algorithm is applied to calculate the shortest path from the user\u27s position to the selected accommodation. This study involved 21 accommodations as study objects. The results of the analysis showed that Hotel Diana obtained the highest value of 0.08873, so it was recommended as the main accommodation. The shortest distance from the Faculty of Engineering, Malikussaleh University to Hotel Diana is 11.53857 km. These results indicate that the combination of the AHP method and the Dijkstra algorithm is effective in supporting location-based decision making, as well as making it easier for users to determine appropriate accommodation and the fastest route efficiently.This study aims to develop a web-based Geographic Information System (GIS) to provide recommendations for the best accommodation and the fastest route to the accommodation location in Lhokseumawe City. The Analytical Hierarchy Process (AHP) method is used to determine the priority of accommodation based on five main criteria, namely price, public facilities, cleanliness, security, and year founded. The Dijkstra algorithm is applied to calculate the shortest path from the user\u27s position to the selected accommodation. This study involved 21 accommodations as study objects. The results of the analysis showed that Hotel Diana obtained the highest value of 0.08873, so it was recommended as the main accommodation. The shortest distance from the Faculty of Engineering, Malikussaleh University to Hotel Diana is 11.53857 km. These results indicate that the combination of the AHP method and the Dijkstra algorithm is effective in supporting location-based decision making, as well as making it easier for users to determine appropriate accommodation and the fastest route efficiently
A Smart Recommendation System for Crop Seed Selection Using Gradient Boosting Based on Environmental and Geospatial Data
No full textThe selection of appropriate crop seeds is a critical factor in enhancing agricultural productivity. Nevertheless, farmers frequently face challenges when trying to determine which crop seeds match the unique features of their surrounding environment and geographic location. To address this, the study introduces a smart recommendation model that leverages real-time environmental measurements alongside vital geographical characteristics to support informed seed selection. The environmental features include temperature, humidity, and rainfall, while the geographical attributes encompass nitrogen, phosphorus, and potassium content. A Gradient Boosting classification algorithm is employed to model the relationships between these features and the optimal crop seed types, based on a labeled dataset. Experimental results demonstrate that the model achieves strong classification performance, indicating its effectiveness in delivering accurate and context-specific seed recommendations. The proposed system highlights the potential of data-driven approaches in supporting agricultural decision-making and can be further integrated into smart farming platforms to optimize crop planning and seed selection, ultimately contributing to improved agricultural outcomes
Hypertension Risk Prediction Using Stacking Ensemble of CatBoost, XGBoost, and LightGBM: A Machine Learning Approach
No full textHypertension is a leading cause of cardiovascular diseases, chronic kidney failure, and strokes, affecting millions worldwide. Early detection and accurate risk prediction are crucial for effective management and prevention. This study aims to evaluate and compare the performance of different algorithms for predicting hypertension risk using a stacking ensemble approach. The model combines three gradient boosting algorithms XGBoost, LightGBM, and CatBoost as base learners, with Logistic Regression as the meta learner. The dataset, sourced from Kaggle, contains 4,240 instances with demographic and clinical attributes relevant to hypertension. The preprocessing steps included imputing missing values using the median, removing residual null entries, and addressing class imbalance through the SMOTE algorithm. Data were divided into 80% for training and 20% for testing. The evaluation showed that the stacking ensemble model achieved an overall accuracy of 92,65%, with precision, recall, and F1-scores consistently reaching 0.92 for both classes. The confusion matrix revealed minimal misclassification, indicating the model’s strong ability to differentiate between low and high risk individuals. These results emphasize that the primary goal of this research is to identify which algorithm provides the best performance for hypertension risk prediction. By evaluating and comparing different models, this study offers insights into choosing the most effective algorithm for clinical decision-making and early detection strategies
Enhancing Clustering Accuracy Using K-Means with Seeds Optimization
No full textIn this study, the development of the Mean-based method proposed by Goyal and Kumar will be carried out by changing the initial cluster center determination step, which was originally based on the origin point O (0,0), to be replaced with the arithmetic mean. To assess the performance of the proposed method, it will be compared with the Global K-means method and the Mean-based K-means method. In this study, the performance of these methods will be measured using the Davies-Bouldin Index, and the significance of the proposed method will be measured using the Friedman Test. This study proposes a method of Improving K-Means Performance through Initial Center Optimization based on Second Global Average for Clustering Osteoporosis Diagnosis of lifestyle factors. Evaluation of K-Means performance through Initial Center Optimization based on Second Global Average with DBI measurements. The targeted experimental results of this study include improving the performance of K-means optimized through the initial center based on Second Global Average. From the results of nine experiments with the number of clusters [2,3,4,5,6], it can be seen that the method proposed in this study has the same superior performance compared to the Mean Based method and compared to the Global K-means method
Comparative Sentiment Analysis on Mobile JKN Application Using Logistic Regression with SMOTE Based Statistical Feature Selection
No full textThis study investigates public sentiment on the Mobile JKN application using Logistic Regression enhanced with SMOTE-based statistical feature selection. Unlike prior works that relied solely on conventional feature combinations such as TF-IDF or Word2Vec, this research performs a comparative evaluation of three statistical feature selection techniques: Recursive Feature Elimination (RFE), Chi-Square, and Mutual Information, under both TF-IDF and Word2Vec representations in a low-resource Indonesian language setting. The dataset consists of 2,382 user reviews from the Google Play Store, balanced using SMOTE to mitigate class imbalance. The best configuration, TF-IDF combined with Mutual Information, achieved an accuracy of 73.38% and an F1-score of 50%, indicating a moderate yet consistent performance. A confusion matrix-based error analysis revealed that most misclassifications occurred between neutral and negative classes due to semantic overlap. The relatively low F1-score highlights challenges in sentiment separability, while the superior performance of Mutual Information demonstrates its ability to capture discriminative linguistic features. The superior performance of Mutual Information is attributed to its ability to capture non-linear dependencies between features and sentiment labels, yielding richer discriminative information compared to Chi-Square or RFE. This research establishes a comparative methodological framework that integrates feature selection and data balancing techniques, providing interpretable sentiment classification insights for under-resourced language settings
Impact of SMOTE and ADASYN on Class Imbalance in Metabolic Syndrome Classification Using Random Forest Algorithm
No full textMetabolic Syndrome is a collection of medical conditions that can increase the risk of stroke, cardiovascular disease, and type 2 diabetes. Early detection of this condition requires a machine learning model capable of accurate classification to support timely treatment. However, class imbalance in data often hampers the performance of classification algorithms, particularly in recognizing minority classes, namely individuals diagnosed with Metabolic Syndrome. This study aims to analyze the effect of applying the SMOTE and ADASYN data balancing techniques in classifying Metabolic Syndrome using the Random Forest algorithm. These algorithms were chosen for their ability to produce accurate predictions, although their performance can decline when faced with imbalanced class distributions. The results showed that the model without data balancing techniques achieved 86% accuracy with a minority class recall of 75%. The application of SMOTE increased accuracy to 91% and recall to 93%, while ADASYN achieved 92% accuracy and a minority class recall of 95%. These findings indicate that the ADASYN technique combined with the Random Forest algorithm provides significant performance improvements in the classification of Metabolic Syndrome on imbalanced data
Comparative Study of Logistic Regression, Random Forest, and XGBoost for Bank Loan Approval Classification
No full textBank loan approval plays a vital role in ensuring financial institutions can minimize credit risk while supporting economic growth. Default prediction is a crucial aspect of banking credit risk management. This study compares three machine learning algorithms Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost) to classify bank loan approvals using a combination of application, previous application, and bureau datasets. The workflow includes data merging, cleaning, missing value imputation, handling unknown values, feature engineering (such as converting day-based variables into years, calculating total submitted documents, income-to-annuity ratio, and employment-to-income ratio), encoding (label and one-hot), scaling (min-max normalization), feature selection based on correlation analysis, handling class imbalance with SMOTE, as well as modeling and evaluation using Accuracy, Precision, Recall, F1-score, and AUC. The results show that Logistic Regression yields the highest AUC of 0.741498, outperforming Random Forest (0.713758) and XGBoost (0.715944). From a business perspective, implementing the best model reduced the Loss Given Default (LGD) by 39.77 %, from 1,026,944,185.50. This finding confirms that simpler models remain competitive on imbalanced datasets when supported by appropriate preprocessing and balancing strategies.Bank loan approval plays a vital role in ensuring financial institutions can minimize credit risk while supporting economic growth. Default prediction is a crucial aspect of banking credit risk management. This study compares three machine learning algorithms Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost) to classify bank loan approvals using a combination of application, previous application, and bureau datasets. The workflow includes data merging, cleaning, missing value imputation, handling unknown values, feature engineering (such as converting day-based variables into years, calculating total submitted documents, income-to-annuity ratio, and employment-to-income ratio), encoding (label and one-hot), scaling (min-max normalization), feature selection based on correlation analysis, handling class imbalance with SMOTE, as well as modeling and evaluation using Accuracy, Precision, Recall, F1-score, and AUC. The results show that Logistic Regression yields the highest AUC of 0.741498, outperforming Random Forest (0.713758) and XGBoost (0.715944). From a business perspective, implementing the best model reduced the Loss Given Default (LGD) by 39.77 %, from 1,026,944,185.50. This finding confirms that simpler models remain competitive on imbalanced datasets when supported by appropriate preprocessing and balancing strategies