BSBagus Sartono

Bagus Sartono

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Bagus Sartono has a total of 184 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    EVALUATION AND DEVELOPMENT OF MACHINE-LEARNING-BASED COMPLIANCE SCORING MODELS FOR ALCOHOLIC BEVERAGES EXCISE TAXPAYER IN INDONESIA WITH INNOVATIVE ENCODING STRATEGY

    by Boy Riansyah, Bagus Sartono, Farit M Afendi

    The transformation of categorical variables as part of the feature engineering process is a critical stage in classification modeling because it determines the quality of the information representation used throughout model development. This issue is particularly relevant to the development of a compliance scoring model for excise taxpayers producing or importing alcoholic beverages in Indonesia, where all predictor variables are categorical. The compliance scoring model currently implemented by the Directorate General of Customs and Excise (DGCE) is based on a manually constructed scorecard in which the categorization of predictor variables is determined subjectively. Moreover, the existing model has not been empirically validated to represent the probability of compliance violations. This study aims to evaluate the current excise compliance scoring model while developing data-driven alternatives using logistic regression and the Categorical Boosting (CatBoost) algorithm with innovative categorical encoding strategies. Two encoding methods are compared: Weight of Evidence (WoE), representing a conventional statistical approach, and Ordered Target (OT), which is adapted from the CatBoost encoding mechanism. All models are validated using repeated k-fold cross-validation and evaluated based on the Area Under the Receiver Operating Characteristic Curve (AUC), the Kolmogorov–Smirnov (KS) statistic, and several additional performance metrics. The results demonstrate that OT encoding consistently outperforms WoE across all logistic regression scenarios, increasing the KS statistic from 0.27 for the existing compliance scoring model to 0.34 with WoE and 0.42 with OT. Furthermore, the CatBoost model achieves superior predictive performance compared with both logistic regression models (AUC = 0.715; KS = 0.471), indicating stronger discriminative ability in identifying compliance violations. These findings demonstrate that integrating target-based encoding with boosting algorithms produces a more adaptive and accurate compliance scoring model, providing a robust analytical framework to support risk-based excise supervision in Indonesia.

  • Open Access

    Gamification and Behavioral Participation in Asynchronous Distance Tutorials: An Exploratory Comparative Study in Open Distance Higher Education

    by Dewi Juliah Ratnaningsih, Pismia Sylvi, Bagus Sartono, Timbul Pardede, Sri Maulidia Permatasari

    Student engagement remains a persistent challenge in open and distance higher education (ODHE), particularly within asynchronous online tutorials where behavioral participation—including attendance, content engagement, and collaborative activity—declines progressively across tutorial sessions. Gamification, the integration of game design elements such as points, badges, levels, and leaderboards into non-game learning environments, has been proposed as an instructional design strategy for sustaining engagement. However, empirical evidence on how varying degrees of gamification comprehensiveness relate to multiple behavioral participation indicators in authentic asynchronous ODHE settings remains limited. This study presents an exploratory comparative analysis of LMS-based behavioral participation across three intact tutorial classes assigned to three instructional conditions: non-gamified (control), partially gamified (points and levels only), and fully gamified (points, levels, badges, leaderboard, and progress indicators). The gamification intervention was implemented through the Moodle Level Up plugin within a compulsory undergraduate statistics course at an open distance university in Indonesia. Five behavioral participation indicators were operationalized from LMS activity log data—attendance (Hadir button click), reading activity, discussion forum participation, formative quiz completion, and assignment submission—across eight tutorial sessions. The same tutor delivered all three classes using identical instructional materials and assessment rubrics. Descriptive analysis revealed consistent participation gradients across all five indicators: the fully gamified class recorded the highest mean participation rates, followed by the partially gamified class, and the non-gamified class. Mean differences between the fully gamified and non-gamified classes ranged from 15.62 to 33.33 percentage points across indicators. These patterns were consistent across all eight tutorial sessions, suggesting sustained rather than novelty-driven differences. Findings are discussed in relation to motivational theory, alternative explanations, and implications for instructional design in ODHE. Limitations of the exploratory comparative design, including class-level confounding and the absence of individual-level data, are acknowledged.

  • Open Access

    Machine Learning Approaches for Household Smoking Classification: A Comparison between Encoded Logistic Regression and XGBoost

    by Jasmin Nur Hanifa, Boy Riansyah, Aulia Rizki Firdawanti, Bagus Sartono, Wildatul Maulidiyah, Felasofa Rahmatanti

    Smoking behavior within households is a significant public health issue, particularly in densely populated regions such as West Java. This study aims to compare the classification performance of Logistic Regression and XGBoost in identifying household smoking status under several feature representation schemes, including baseline models, One-Hot Encoding, and Weight of Evidence (WoE). The analysis used 23.854 household observations from 2024 National Socioeconomic Survey (SUSENAS) in West Java after preprocessing. Model performance was evaluated using 5-fold stratified cross-validation based on accuracy, sensitivity, specificity, and Area Under the Curve (AUC) metrics. Among the evaluated configurations, XGBoost with WoE Encoding achived the highest observed accuracy (0.739), sensitivity (0.782), and AUC (0.797), although the differences in AUC across models were relatively small. Gain-based feature importance identified the Food–Non-Food Expenditure Ratio (X1) as the most influential predictor. The dominance of this expenditure ratio confirms that household expenditure structure, as an indicator of economic condition and welfare, plays a crucial role in distinguishing smoking households. These findings support the hypothesis that poorer households tend to have a higher probability of smoking.

  • Open Access

    A Hybrid ChemBERTa–Morgan XGBoost Ensemble for Computational Multi-Endpoint Toxicity Prioritization of Blumea balsamifera Compounds

    by Kusnaeni Kusnaeni, Agus Mohamad Soleh, Bagus Sartono, Taufik Adhi Prasetya Wardana, Fitriagustiani Fitriagustiani, Muhammad Fadhil Nurahmad, Muhammad Rahmatullah, Muhammad Rizqy Firmansyah, Afif Budi Andy B

    Biological investigations of Blumea balsamifera remain dominated by extract or fraction-based testing; consequently, the activities of individual compounds across various biological mechanisms have not been consistently characterized. This study developed a computational framework to prioritize 31 Blumea balsamifera compounds based on their predicted activities across 12 Tox21 endpoints. The framework integrates Morgan fingerprints, RDKit descriptors, and ChemBERTa embeddings through two endpoint-specific XGBoost branches combined using an equal-weight soft voting ensemble of calibrated probabilities. The models were developed using 7,820 Tox21 structures after three structures identical to the target compounds were excluded. Evaluation was performed using repeated five-fold scaffold cross-validation, Platt scaling, endpoint-specific thresholds, and applicability domain analysis. The ensemble achieved mean ROC-AUC, average precision, balanced accuracy, and MCC values of 0.784±0.063, 0.372±0.130, 0.712±0.052, and 0.273±0.103, respectively. Luteolin-7-methyl-ether ranked highest, with a score of 37.10 and predicted activity at 11 endpoints. This framework supports the selection of compounds for more targeted biological validation.• Develops a hybrid framework for computational prioritization of toxicity-related activity across 12 Tox21 endpoints.• Integrates Morgan fingerprints, RDKit descriptors, and ChemBERTa embeddings into a calibrated XGBoost ensemble.• Identifies luteolin-7-methyl ether as the highest priority candidate for multi-endpoint biological validation.

  • Open Access

    Meta-stacking models for electricity load forecasting in West Java

    by Denanda Aufadlan Tsaqif, Bagus Sartono, Hari Wijayanto

    Indonesia’s electricity demand continues to increase due to population growth, urbanization, and industrial expansion, therefore making accurate load forecasting is essential to maintain supply-demand balance. However, electrical load demand in West Java has a complex pattern (seasonality, nonlinear behavior, weather variability, and holiday effects), which motivates the use of a meta-stacking approach to effectively capture such complexity. Previous research shows that meta-stacking outperforms individual models, but it fails to capture sudden changes and its performance consistency remains unclear. Therefore, this study proposes a meta-stacking framework for daily electricity load forecasting in West Java (2006-2023) that includes weather and holiday variables by combining CNN-BiLSTM, CNN-BiGRU, and Windowed-XGBoost forecasts through linear regression and evaluates its performance across five data-splitting scenarios and nine forecast horizons, which represents the main novelty in this research. Meta stacking shows strong generalization across scenarios and strong long-term forecasting performance across horizons, while consistently providing a balanced trade-off between MAPE and trend accuracy, where the model trained on the longest historical dataset achieves the best performance with 1.89% MAPE and 86% trend accuracy. The proposed approach successfully captures seasonal and holiday-related load patterns, indicating its potential to support PLN in improving demand planning and operational decision making.

  • Open Access

    EFFECTIVENESS OF DIMENSIONALITY REDUCTION METHODS ON DATA WITH NON-LINEAR RELATIONSHIPS

    by Lukmanul Hakim, Asep Saefuddin, Kusman Sadik, Anwar Fitrianto, Bagus Sartono

    The phenomenon of big data presents distinct challenges in the analysis process, especially when the data contains a very large number of variables. High complexity, potential redundancy, and the risk of overfitting are major issues that must be addressed through dimensionality reduction techniques. Principal Component Analysis (PCA) is a common method effective for data with linear relationships but has limitations in identifying nonlinear patterns. This research aims to improve performance of classification by introducing autoencoder for dealing with nonlinear relationship, data noise, missing values, outliers, and data with various scales. This study employs a quantitative approach through analysis of simulated and empirical data in the form of the Village Development Index from the Central Statistics Agency, which contains variables with various measurement scales. Both dimensionality reduction methods—PCA and neural network-based autoencoders—are tested across various data scenarios. The evaluation is conducted based on their effectiveness in preserving data structure, as well as the Mean Squared Error (MSE) values in the reconstruction process. The results indicate that PCA excels in computational efficiency and accuracy for data with linear relationships. In contrast, the autoencoder demonstrates superior performance in detecting nonlinear patterns, achieving lower Mean Squared Error (MSE) values with stable MSE standard deviations. Additionally, the autoencoder proves to be more robust in handling missing values and outliers compared to PCA. The selection of dimensionality reduction methods highly depends on the characteristics of the analyzed data. Autoencoders represent a superior alternative for handling complex and nonlinear data, although they require model parameter tuning. Further research is recommended to explore the influence of network architecture and training strategies of autoencoders on dimensionality reduction performance.

  • Open Access

    COMPARISON OF MIXED EFFECT REGRESSION TREE (MERT) AND LINEAR MIXED MODEL (LMM) FOR CLUSTERED DATA ON CASE STUDY HOUSEHOLD POVERTY IN WEST JAVA PROVINCE

    by Nur Fitriyani Sahamony, Asysta Amalia Pasaribu, Bagus Sartono, Khairil Anwar Notodiputro

    This study compares the performance of the Linear Mixed Model (LMM) and the Mixed Effect Regression Tree (MERT) in analyzing the determinants of household consumption expenditure in West Java Province. The LMM integrates fixed and random effects to account for both individual and regional variations, while MERT extends this approach by incorporating a regression tree framework to capture nonlinear relationships and complex interactions among socio-economic variables. Using data from the 2023 National Socioeconomic Survey (SUSENAS), household consumption expenditure is modeled as an indicator of poverty. The results show that key determinants across both models include the gender and age of the household head, highest educational attainment, household size, land and car ownership, and welfare card ownership. Education and asset ownership consistently emerge as major factors influencing household welfare. The MERT model demonstrates superior predictive performance, with lower RMSE and MAE values compared to the LMM, while offering greater interpretability by identifying specific household profiles. Female-headed households with higher education and no car ownership tend to have higher expenditure in the high-income group, whereas female-headed households with welfare cards remain vulnerable in the low-expenditure group. From a policy perspective, these findings highlight the importance of improving educational access, enhancing asset ownership, and strengthening targeted social protection for vulnerable groups. Overall, while both models contribute valuable insights, the MERT model provides a more flexible and powerful framework for identifying and interpreting the determinants of household welfare in West Java.

  • Enhancing health survey data modeling through mixed-effects machine learning: A comparative study

    by Siskarossa Ika Oktora, Dariani Matualage, Khairil Anwar Notodiputro, Bagus Sartono

    Epidemiological studies frequently utilize hierarchical data structures, such as regional variations. Central obesity, commonly measured by waist circumference, is a key predictor of cardiovascular disease, one of the leading global causes of mortality. Intra-cluster correlations arising from lifestyle disparities and unequal access to healthy resources necessitate statistical models that account for both between- and within-cluster variability. Linear Mixed-Effects Models are often used for this purpose, but they may fall short in capturing nonlinearities and complex interactions among predictors. To address these limitations, tree-based extensions such as the Mixed-Effects Regression Tree and Mixed-Effects Random Forest have been introduced. MERF integrates Random Forests into the mixed-effects framework to enhance flexibility and predictive power. This study evaluates and compares the performance of LMM, MERT, and MERF in modeling waist circumference as a proxy for central obesity, using Indonesia's 2018 Basic Health Survey for West Java Province. Two modeling strategies were applied: one using full province-wide data, and another stratifying the regions into three risk categories (high, moderate, and low) based on proximity to the Special Capital Region of Jakarta. Moreover, separate models were developed for each gender to examine any difference in the contributing or influential variables for waist circumference between males and females. All models included regency/city as a random effect. Results show that MERF outperforms LMM in predictive accuracy and performs comparably to MERT, highlighting its potential for improving the modeling of clustered health survey data. Linear-based approaches, such as LMM and simple tree-based models, specifically exhibited better predictive performance than MERF for the male model. The findings support the use of mixed-effects machine learning approaches in enhancing the quality of official health statistics and informing targeted public health interventions.


  • Open Access

    TREE-BASED MIXED EFFECTS MODELING OF TEACHER CERTIFICATION OUTCOMES IN MADRASAH ALIYAH: A COMPARATIVE STUDY OF GLMM TREES AND GMET

    by Dodi Irawan Syarip, Khairil Anwar Notodiputro, Bagus Sartono

    The Teacher Professional Education program, or “Pendidikan Profesi Guru” (PPG), is a continuing education program designed for prospective or in-service teachers to obtain a teaching certificate. PPG is a priority program of the Ministry of Religious Affairs in providing competent and professional madrasah teachers. This study is expected to identify the challenges encountered in the implementation of the Madrasah teacher certification program and provide valuable input to enhance the success rate of Madrasah Aliyah teachers in the PPG program. The main objective of this study is to find the most appropriate tree-based mixed effects model to analyze the effectiveness of PPG for Madrasah Aliyah teachers in 2022. This study applies two tree-based mixed effects modeling methods: generalized linear mixed model trees (GLMM trees) and generalized mixed effects trees (GMET). Both methods model variability across subjects as a random effect. Based on the performance indices measurement results, the GMET model shows superiority over the GLMM trees model. The GMET model has an accuracy index of 0.7653, higher than the GLMM trees model of 0.7306. Substantively, teachers of English and Indonesian Language exhibit higher probabilities of passing than those of other subjects, whereas Arabic and Islamic Cultural History have the lowest estimated probabilities of success. Analysis of the variable importance from both models indicates that teachers’ age is the most influential predictor of PPG graduation among Madrasah Aliyah teachers. Based on these findings, to improve the effectiveness of PPG implementation for madrasah Aliyah teachers, policymakers at the Ministry of Religious Affairs are advised to implement a structured coaching and mentoring program for prospective PPG participants, with a special emphasis on support for senior teachers specializing in Arabic and Islamic Cultural History.

  • Open Access

    COMPARISON OF SIMPLEX AND NELDER-MEAD OPTIMIZATION METHODS IN QUANTILE REGRESSION FOR BOGOR CITY RAINFALL ANALYSIS

    by Salsa Rifda Erira, Delia Fitri Audina, Meriza Immanuela Virgie, ⁠bUlan Cahyani Suhaeri, Muhammad Fatih Abyan, Akbar Rizki, Bagus Sartono

    Predicting extreme rainfall is crucial for supporting planning in the agricultural sector, infrastructure development, and disaster mitigation in the city of Bogor. However, the asymmetric distribution of daily rainfall and the presence of outliers make linear regression methods less suitable. Quantile regression offers an alternative that captures the influence of explanatory variables across different parts of the data distribution, particularly in the extreme regions. This study compares the Simplex and Nelder-Mead methods for estimating quantile regression parameters on extreme rainfall data in Bogor. Daily rainfall data were obtained from the West Java BMKG Climate Station for the period from May 2024 to April 2025, comprising 365 observations, with four explanatory variables: average temperature, average humidity, sunshine duration, and average wind speed. Modeling was conducted at the 0.75, 0.85, and 0.95 quantiles to represent extreme rainfall. The results show that the Simplex method outperformed Nelder-Mead, as indicated by lower Pinball Loss and Mean Absolute Error (MAE) values at most quantiles. Humidity and average wind speed had a significantly positive effect on extreme rainfall intensity, while average temperature had a negative effect. Sunshine duration showed less consistent effects. Overall, the Simplex method is recommended for quantile regression optimization in extreme rainfall data due to its greater stability and accuracy in generating model parameters. However, this study is limited by the number of explanatory variables and the relatively short observation period. Incorporating additional variables such as air pressure, ENSO index, or topographical data, along with extending the observation period, could improve model accuracy and generalizability in future research.