Publication Search

This research benchmarks multiple machine learning (ML) algorithms for large-scale loan default prediction using a real-world dataset of 255,000 borrower records, where default cases represent only ~9–12% of total observations. The study addresses the persistent gap in comparative analyses of ML models that balance predictive accuracy, interpretability, and computational efficiency for credit risk assessment. Six algorithmic families were evaluated Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost, Artificial Neural Networks (ANN), and Stacked Ensemble—using standardized preprocessing, hybrid imbalance handling (SMOTE, class weighting, under-sampling), and comprehensive evaluation metrics (AUC, F1, Recall, Precision, PR-AUC, and Brier Score). Empirical results show Logistic Regression achieved the highest AUC of 0.732, outperforming nonlinear models under the baseline configuration, while LightGBM attained perfect recall (1.0) but low precision (0.116), indicating over-prediction of defaults. Gradient boosting models demonstrated robust calibration (Brier ≈ 0.114–0.116) and the best computational efficiency, with LightGBM showing the fastest training and lowest memory use. CatBoost exhibited strong recall but the slowest computation, and ANN underperformed on tabular data (AUC ≈ 0.56). The Stacked Ensemble delivered balanced results with AUC = 0.664 and improved overall stability. These findings confirm that boosting-based models, particularly LightGBM and CatBoost, offer superior scalability and calibration, whereas Logistic Regression remains a valuable interpretable baseline. The study concludes that effective default prediction requires integrating rebalancing, calibration, and threshold optimization to enhance recall and operational deployment reliability in large-scale credit ecosystems.

This study aims to estimate coal resources using the kriging method at PT. Inti Bara Perdana, Taba Penanjung, Bengkulu. Geostatistical analysis was applied to evaluate spatial variability of coal quality parameters, including total moisture (TM), inherent moisture (IM), ash content (ASH), volatile matter (VM), fixed carbon (FC), and calorific value (CV). The research utilized drilling data with an average spacing of 80 meters. Variogram modeling was conducted using SGeMS software, employing spherical models to determine nugget, sill, and range parameters. The results show a low nugget effect (0%) indicating strong spatial continuity. The obtained ranges vary between 180–1296 meters depending on the parameter. Blo./ck kriging estimation was performed using block dimensions of 25 × 25 × 8 meters and a coal density of 1.3 ton/m³. The total estimated coal resource up to an elevation of 230 m is 253,500 tons. These findings demonstrate that kriging provides reliable resource estimation and supports mine planning optimization. This research contributes to improving coal resource evaluation accuracy and operational decision-making in open-pit mining.

Drug stockouts in hospitals pose significant risks to service quality, patient safety, and operational efficiency. This study aimed to analyze how drug demand planning and procurement processes at Johar Baru Regional General Hospital contribute to stockout occurrences and to develop data-driven recommendations based on supply chain management principles. A qualitative descriptive design was employed using data triangulation. Data were collected through in-depth interviews with the Head of the Pharmacy Installation, procurement staff, and warehouse pharmacists, complemented by direct observation and analysis of 2024 planning and procurement documents. Thematic analysis was conducted with the support of NVivo software to identify patterns and relationships among key variables, including drug demand planning, procurement, and inventory management.Findings reveal that stockouts stem from interconnected weaknesses in planning accuracy, procurement coordination, and inventory control systems. Effective stock management depends not only on increasing supply but also on improving data quality, integrating inventory information systems with operational workflows, and enhancing cross-functional collaboration. Recommended strategies include implementing a minimum stock alert system integrated with the Hospital Management Information System (HMIS), strengthening standard operating procedures for stockout response and procurement confirmation, improving integration between HMIS, the National Formulary, and budgeting systems, and applying consumption based planning methods combined with ABC VEN analysis to optimize inventory control.

This research try to examine how investment choices in property and real estate sector businesses listed on IDX (2020-2024) are impacted by the cost of debt and equity.For the accuracy of the analysis, this study also uses firm size and profitability as controler. Purposive sampling was used in the sampling process, which produced seven qualifying organizations with a total of 35 observations over a five-year period. Panel data regression was used for data analysis, and the Common Effect Model was shown to be the best estimation model. The findings show that investment decisions are significantly influenced by firm size, profitability, cost of debt, and cost of equity all at the same time. Nonetheless, investment choices are not much impacted by the cost of debt. Similarly, it has been demonstrated that the cost of equity has no appreciable effect on the capital expenditures of the businesses. Firm size has a favorable and substantial impact, making it the main motivator for investment activity in the real estate industry. During the study period, investment decisions were not significantly impacted by profitability. These results show that, especially in the post-pandemic economic recovery era, asset capacity and economies of scale are more important for the viability of real investment projects for property firms on the IDX than yearly variations in capital costs.

English speaking skill is a crucial competence for university students, particularly those involved in student organizations focusing on language development. Members of English clubs are expected to actively use English in academic and organizational contexts; however, many still face challenges related to pronunciation accuracy, fluency, limited active vocabulary, and low confidence. These challenges are often caused by limited structured speaking practice and insufficient integration between digital learning tools and communicative pedagogy.This community service program aimed to enhance the English-speaking skills of members of the Bhamada English Club at Universitas Bhamada Slawi through the utilization of the Duolingo application integrated with a Task-Based Learning (TBL) approach. The program was implemented through three stages: preparation, implementation, and evaluation. Activities included needs analysis, speaking pre-test, guided Duolingo practice focusing on pronunciation and speaking features, task-based speaking activities, and post-test evaluation.The results demonstrated improvements in pronunciation accuracy, speaking fluency, and participants’ confidence in using English orally. The integration of Duolingo as a source of comprehensible input and Task-Based Learning as a communicative output strategy proved effective in enhancing speaking skills. This program indicates that technology-supported task-based instruction can serve as an effective and sustainable model for improving English speaking skills among university students.

The computer laboratory is an essential facility in higher education that requires efficient management of usage and environmental conditions to support the teaching and learning process. However, laboratory management at the Kalimantan Business and Technology Institute is still carried out manually, including scheduling, room condition monitoring, and report creation, which is prone to errors and time-consuming. This study aims to develop an Internet of Things (IoT)-based laboratory monitoring system prototype to improve the effectiveness of computer laboratory management. The approach used is Research and Development (R&D) with a prototype development model, allowing for design adjustments based on user feedback iteratively. Data were collected through observations, interviews, and document studies related to laboratory conditions and analyzed to determine the main system features, such as temperature and humidity monitoring, scheduling, and report generation. The results of the study show that the developed prototype can structure the laboratory workflow, provide real-time monitoring, facilitate schedule management, and simplify report preparation. This prototype is expected to serve as a foundation for developing a more comprehensive application, improving data accuracy, time efficiency, and the quality of laboratory management.

The computer laboratory is an essential facility in higher education that requires efficient management of usage and environmental conditions to support the teaching and learning process. However, laboratory management at the Kalimantan Business and Technology Institute is still carried out manually, including scheduling, room condition monitoring, and report creation, which is prone to errors and time-consuming. This study aims to develop an Internet of Things (IoT)-based laboratory monitoring system prototype to improve the effectiveness of computer laboratory management. The approach used is Research and Development (R&D) with a prototype development model, allowing for design adjustments based on user feedback iteratively. Data were collected through observations, interviews, and document studies related to laboratory conditions and analyzed to determine the main system features, such as temperature and humidity monitoring, scheduling, and report generation. The results of the study show that the developed prototype can structure the laboratory workflow, provide real-time monitoring, facilitate schedule management, and simplify report preparation. This prototype is expected to serve as a foundation for developing a more comprehensive application, improving data accuracy, time efficiency, and the quality of laboratory management.

Nabire Regency in Central Papua Province confronts significant geographical complexities in transportation system development, characterized by hilly topography and limited road infrastructure that impacts community mobility efficiency. This research analyzes road route optimization through Geographic Information System (GIS) approach using QGIS 3.28.3 software with QNEAT3 plugin for network analysis. The research methodology applies network analysis considering multiple parameters including travel distance, travel time, road surface conditions, slope gradients based on Digital Elevation Model (DEM), and impedance factors according to infrastructure quality. Spatial data were obtained from Geospatial Information Agency (BIG) and OpenStreetMap (OSM) with analysis focusing on 8 origin-destination pairs representing critical movements to health facilities, education centers, and economic hubs. Analysis results identified alternative routes demonstrating 12.5-20% travel time efficiency compared to existing routes, with an average optimization of 15.7%. Field validation confirmed model prediction accuracy with error rates below 8%. These research findings provide strategic recommendations for local government in sustainable transportation infrastructure planning that can enhance regional accessibility and support local economic development in the Central Papua region.

Prompt injection has emerged as a critical security threat for Large Language Models (LLMs), exploiting their inability to separate instructions from data within application contexts reliably. This paper provides a structured review of current attack vectors, including direct and indirect prompt injection, and highlights the limitations of existing defenses, with particular attention to the fragility of Known-Answer Detection (KAD) against adaptive attacks such as DataFlip. To address these gaps, we propose a novel, hybrid, multi-layered detection framework that operates in real-time. The architecture integrates heuristic pre-filtering for rapid elimination of obvious threats, semantic analysis using fine-tuned transformer embeddings for detecting obfuscated prompts, and behavioral pattern recognition to capture subtle manipulations that evade earlier layers. Our hybrid model achieved an accuracy of 0.974, precision of 1.000, recall of 0.950, and an F1 score of 0.974, indicating strong and balanced detection performance. Unlike prior siloed defenses, the framework proposes coverage across input, semantic, and behavioral dimensions. This layered approach offers a resilient and practical defense, advancing the state of security for LLM-integrated applications.

The increasing complexity and scale of modern network traffic driven by IoT and cloud-based infrastructures have made accurate intrusion detection a critical challenge. Conventional network intrusion detection systems (NIDS) and many deep learning–based approaches struggle to reliably detect minority and stealthy attacks due to severe class imbalance and limited discrimination of subtle traffic patterns. To address these limitations, this study proposes a hybrid CNN–RBF–Attention framework for network intrusion detection. The proposed model integrates three complementary components: (i) a convolutional neural network for hierarchical feature extraction from network flow data, (ii) a radial basis function (RBF) network for localized nonlinear classification using prototype-based decision regions, and (iii) an attention mechanism that adaptively weights RBF activations to emphasize discriminative traffic patterns. SMOTE is applied exclusively to the training data to mitigate class imbalance. The framework is evaluated on the widely used CICIDS2017 and CICIDS2018 benchmark datasets in both binary and multiclass settings, using recall, precision, F1-score, confusion matrices, and ROC analysis. Experimental results demonstrate that the proposed hybrid model consistently outperforms standalone CNN and RBF baselines, particularly in terms of recall and F1-score. On the CICIDS2018 dataset, the model achieves 99.81% accuracy and 99.81% F1-score in binary classification, and 99.54% accuracy and 99.54% F1-score in multiclass classification. On CICIDS2017, it achieves 98.12% accuracy and 98.12% F1-score in binary classification, and 98.92% accuracy and 98.92% F1-score in multiclass classification. Confusion matrix and ROC analyses further show strong class separability and reliable performance in low–false-positive-rate regions, which is critical for real-world IDS deployment. These results confirm that combining deep hierarchical feature learning, localized prototype-based classification, and attention-guided refinement yields a robust, operationally reliable intrusion detection framework for highly imbalanced network environments.

This study presents a comprehensive comparative analysis of four traditional machine learning algorithms Decision Tree, Random Forest, K-Nearest Neighbors, and Support Vector Machine for Android malware detection using the preprocessed TUANDROMD dataset comprising 4,465 instances and 241 features representing both static and dynamic application characteristics. Motivated by the limitations of conventional signature-based and hybrid detection methods, especially in managing imbalanced datasets and detecting emerging malware variants, the study employed SMOTE to ensure balanced training data and fair model evaluation. The dataset was divided into 80% training and 20% testing subsets, and models were assessed using key performance metrics including accuracy, precision, recall, F1-score, and ROC AUC. The findings revealed that the proposed Random Forest model outperformed the other classifiers, achieving an accuracy of 0.993, precision of 0.992, recall of 1.000, F1-score of 0.996, and a near-perfect ROC AUC of 0.9998 surpassing state-of-the-art approaches. These results affirm the superior predictive capability, consistency, and robustness of the Random Forest algorithm in Android malware detection. The study concludes that base models, when integrated with class-balancing techniques, provide reliable and efficient malware detection across imbalanced datasets. For future research, the study recommends exploring advanced hybrid or ensemble frameworks that integrate Random Forest with deep learning architectures or other meta-heuristic optimization techniques to further enhance detection accuracy, adaptability, and resilience against rapidly evolving Android malware threats.

The integration of Artificial Intelligence (AI) into precision agriculture has significantly improved plant disease recognition; however, many existing deep learning models remain computationally expensive and feature-redundant, limiting their deployment on low-power and edge devices. To address these limitations, this study proposes a lightweight framework for maize leaf disease recognition based on serial deep feature extraction, dimensionality reduction, and machine-learning–based classification. A pre-trained MobileNetV2 network is employed as a fixed feature extractor to obtain discriminative visual representations, while Principal Component Analysis (PCA) is applied to reduce feature dimensionality by approximately 76%, retaining 95% of the original variance and improving computational efficiency. The compressed features are subsequently classified using a Radial Basis Function Support Vector Machine (RBF-SVM), optimized via grid search and cross-validation. Experiments conducted on a four-class maize leaf disease dataset (Northern Leaf Blight, Common Rust, Gray Leaf Spot, and Healthy), with class imbalance handled during training, demonstrate that the proposed MobileNetV2–PCA–SVM pipeline achieves 97.58% accuracy, 96.60% precision, 96.59% recall, and 96.59% F1-score, outperforming the DenseNet201 + Bayesian-optimized SVM baseline (94.60%, 94.40%, 94.40%, and 94.40%, respectively). This improvement corresponds to a 2.98% accuracy gain, a 55% reduction in error rate, an 86% reduction in model parameters (20.31M to 2.75M), and an 85% reduction in model size (81 MB to 12 MB). These results indicate that the proposed framework provides a compact and efficient solution with strong potential for deployment in resource-constrained agricultural environments.

Private vehicles are a frequently used mode of transportation because they are considered more practical. However, using private vehicles carries several risks, such as traffic accidents due to drivers losing focus on the road due to other activities, such as making calls on smartphones, drinking, or operating the radio. Approximately 90% of accidents are caused by human error. Convolutional Neural Network (CNN) is a type of neural network commonly used on image data. CNN is often used for image classification due to its high performance and accuracy. Therefore, this study aims to analyze the performance of CNN for the classification of distracted driving activities. The results show that the CNN model is able to effectively classify images of distracted driving activities, with an accuracy of approximately 99% across all datasets and across all input image size variations. Furthermore, the results of this study also show that differences in right-hand and left-hand drive datasets do not significantly affect model accuracy. Variations in input image size also do not significantly affect model accuracy, but do affect the training duration.

Improper water management in rice cultivation can lead to water stress, which reduces productivity. Conventional monitoring has limitations on large-scale lands, necessitating more efficient remote sensing technologies. This study aims to develop a water stress identification system for rice plants in the late vegetative phase using multispectral drone imagery integrated with an Artificial neural network (ANN). The research method employs an experimental approach with six water availability levels in Karyamukti Village, Sumedang. Field reference data were obtained through soil moisture sensors converted into Available Water (AW) values. Image processing stages included orthomosaic reconstruction, leaf object segmentation, and transformation of vegetation indices (NDVI, NDRE, GNDVI, etc.) as model inputs. The results show that the ANN model with a four-hidden-layer architecture achieved training and validation accuracies of 94–95%. In the independent testing phase, the model produced an accuracy of 94.60% with an F1-Score of 93.33%. Spatial visualization of the prediction results indicates a consistent water condition distribution across rice plots. In conclusion, the integration of multispectral drones and ANN provides an accurate non-destructive solution for spatial monitoring of water availability in rice plants.

Indramayu mango is a seasonal fruit that is highly favored due to its delicious taste and high nutritional content. However, high mango production is often not supported by adequate post-harvest facilities, particularly in terms of fruit ripeness classification. Currently, mango ripeness classification is still performed manually, which tends to be subjective and inconsistent. To address this issue, this study proposes a ripeness detection system for Indramayu mangoes by integrating the TGS2602 gas sensor and the YOLOv11 algorithm based on image processing. The TGS2602 sensor is used to detect ethylene gas emitted by ripe mangoes, while YOLOv11 is employed for visual image analysis of the fruit. This study aims to evaluate the system’s performance in classifying ripe and unripe mangoes, as well as analyze the integration between the gas sensor and the object detection model. The test results show that the TGS2602 sensor can detect increased ethylene gas concentration in ripe mangoes, while YOLOv11 demonstrates high accuracy in detecting mangoes based on visual images, with precision and recall close to 1.0. The system was also tested under various lighting conditions, including dark environments, and still performed well, although with a slight decrease in accuracy under low-light conditions.

Freshwater fish diseases significantly affect aquaculture productivity and economic sustainability, while accurate visual classification remains challenging due to interclass similarity and image variability. This study presents a comparative evaluation of three deep learning architectures—DenseNet201, ResNet50, and EfficientNetV2-S—using a stepwise optimization strategy combined with Gradient-weighted Class Activation Mapping (Grad-CAM) for freshwater fish disease classification. Models were trained through three phases: baseline, optimized, and fine-tuned. Performance was evaluated using accuracy, precision, recall, F1 score, Matthews correlation coefficient (MCC), Cohen’s kappa, and per-class ROC–AUC. Results show consistent performance improvement across all architectures, with EfficientNetV2-S achieving the highest accuracy (97.14%), followed by ResNet50 (96.11%) and DenseNet201 (94.40%). High ROC–AUC values (>0.98) indicate strong discriminative capability. Grad-CAM analysis confirms that all optimized models focus on biologically relevant lesion regions, enhancing model transparency and reliability.

This study evaluates public sentiment toward Constitutional Court Decision No. 90/PUU-XXI/2023 regarding the age limit for presidential and vice-presidential candidates, a controversial issue closely related to Indonesia’s democratic dynamics. Understanding public opinion on Twitter, as a major platform for political expression, is essential for informing electoral policy formulation. Data were collected using Tweet Harvest through Google Colab and analyzed using the Naïve Bayes algorithm as the primary sentiment classification method, with RapidMiner employed to support and streamline the analytical process. The analysis process included data cleaning, text normalization, stopword removal, manual labeling of 80 tweets as training data, and automatic sentiment classification to identify positive and negative sentiments. From a total of 151 analyzed tweets, 84 (55.63%) were classified as negative and 67 (44.37%) as positive, with the model achieving an accuracy of 66.67%. These findings suggest a tendency toward public opposition to the decision, reflecting dissatisfaction among Twitter users. The study demonstrates that Naïve Bayes is reasonably effective for sentiment classification with limited datasets and provides insights for policymakers in understanding public responses to election-related regulations.

This research develops a machine learning model to classify customer loyalty using the Random Forest algorithm. Customer churn is a critical issue that reduces revenue and increases acquisition costs. A dataset of 50,000 customers from global e-commerce and subscription platforms was processed through data cleaning, imputation, outlier handling, and class balancing with SMOTE. The Random Forest model was built as a baseline and optimized with hyperparameter tuning. Evaluation using accuracy, precision, recall, and F1-score shows that the optimized model achieved 90.81% accuracy and 83.87% F1-score, outperforming previous Naïve Bayes approaches. Feature importance analysis highlights customer service interactions, lifetime value, and demographic factors as key predictors of churn. These findings demonstrate Random Forest’s effectiveness in churn prediction and provide practical insights for customer retention strategies

Open-pit mining activities in Indonesia contribute significantly to the national economy but require stringent monitoring to mitigate environmental degradation. Conventional monitoring methods relying on terrestrial surveys are often constrained by vast coverage areas, high operational costs, and limited field accessibility. This study aims to develop an artificial intelligence model capable of automatically detecting and mapping mining areas to enhance surveillance efficiency. The applied method is Deep Semantic Segmentation utilizing the U-Net Convolutional Neural Network (CNN) architecture. The model was trained using Sentinel-2 satellite imagery, focusing exclusively on Red, Green, and Blue (RGB) spectral channels to replicate human visual perception. Experimental results demonstrate that the proposed model performs reliable segmentation of mining areas, achieving an Accuracy of 93.58% and a Global Intersection over Union (IoU) of 0.8067. These findings indicate that the U-Net architecture can effectively extract spatial features of mines even when utilizing standard visual data. This research contributes to the development of an efficient, cost-effective, and scalable digital monitoring prototype to support innovation in sustainable environmental governance.

The ability to read the Qur’an is a fundamental competence in Islamic education. However, many children still face difficulties in reading the Qur’an accurately and fluently in accordance with tajwid rules. This community service program (Pengabdian kepada Masyarakat/PKM) aimed to improve Qur’anic reading skills of students at Asy-Syakirin Qur’anic Learning Center (TPQ) in Nibung Village, Kapuas Hulu Regency, West Kalimantan. The program employed a Participatory Action Research (PAR) approach, actively involving teachers and students in all stages of the activities. The implementation consisted of planning, action, observation, and reflection stages. During the planning phase, students’ initial reading abilities were identified and grouped into basic (iqra’) and advanced (tahsin) levels, appropriate learning methods were selected, and supporting learning media were prepared. The action phase involved individual and small-group mentoring, guided Qur’an reading with direct correction of pronunciation (makharij al-huruf) and tajwid, application of memorization and dictation (imla’) techniques, and varied teaching approaches through instructor rotation. Evaluation was conducted through pre-tests and post-tests, direct observation of students’ reading performance, and assessment of participation and attendance. The results indicate a significant improvement in students’ Qur’anic reading skills, particularly in fluency, accuracy of pronunciation, and basic understanding of tajwid. This program is expected to serve as a sustainable model for improving Qur’anic literacy in Qur’anic learning institutions.