Each educational institution has designed an academic system with the aim of providing as perfect learning as possible to students. The quality of good students is influenced by various factors, one of which is the available academic system. Previous research has shown that the quality of a student, which can be called academic achievement, can be determined through historical data on the student admission process. This research aims to process one of the admission processes previously implemented in Indonesian state universities using the National Selection for State University Entrance (SNMPTN) data, combined with Cumulative Achievement Index (GPA) data, so that it can be processed using a machine learning model. The algorithm used to create the model is a Supervised Learning Classification algorithm, which includes a Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB). The research was carried out in three schemes based on the percentages of training data and test data. The results obtained show that DT produces the highest accuracy and precision values, with an accuracy value of 0.79 and a precision value of 0.56, respectively. The XGB produces the highest recall and f1-score values, with a recall value of 0.35 and an f1-score value of 0.36. The model with the highest f1-score can be selected as the best model, namely, the model with the XGB algorithm on a 70%-30% train-test data scheme. The resulting model achieved a success rate of 77%.
1 April 2025 in BAREKENG: Jurnal Ilmu Matematika dan Terapan
Stock prices have unstable movements, so forecasting is needed to decide to invest appropriately according to the strategy. Fuzzy Time Series (FTS) uses fuzzy sets to forecast future time series values using historical data. However, interval partitioning in FTS needs to be considered as it can affect the forecasting results. FCM is applied to solve the problem of interval assignment in the universe of discourse. It allows the evaluation of the distribution of historical data and forming intervals of different sizes. Type 2 Fuzzy Time Series (T2FTS) is an extension of FTS to improve forecasting performance and refine fuzzy relationships. This research aims to improve forecasting accuracy using the Fuzzy C-Means (FCM)-T2FTS combination. This research uses daily data on BBRI stock prices from January 2023 to May 2024, with the variables used being close, high, and low prices. The results showed that determining the interval length using unequal length is more efficient than fixed interval length and can improve model performance, demonstrated from the MAPE values of T2FTS and FCM-T2FTS, which are 2.09% and 1.97%, respectively, the difference between the two MAPEs, is 0.12%. Hence, FCM-T2FTS is 12% more efficient than T2FTS. Therefore, FCM-T2FTS can improve forecasting accuracy.
This research aims to optimize forex trading profit automatically using EA but its still keep considering accuracy and drawdown levels. The evaluation system will classify EA performance based on trading market sessions (Sydney, Tokyo, London and New York) to determine the right EA to be used in certain market sessions. This evaluation system is a web-based ELECTRE methods that interact in real-time with EA through web service and are able to present real-time charts performance dashboard using web socket protocol communications. Web applications are programmed using NodeJs. In the testing period, all EAs had been simulated 24 hours in all market sessions for three months, the best EA is valued by its profit, accuracy and drawdown criteria that calculated using web-based ELECTRE method. The ideas of this research are to compare the best EA on testing period with collaboration performances of each best classified EA by market sessions. This research uses three months historical data of EUR/USD as testing period and other 3 months as validation period. As a result, performance of collaboration four best EA classified by market sessions can increase profits percentage consistently in testing and validation periods and keep securing accuracy and drawdown levels.
Chronic hepatitis continues to pose a significant global health challenge, frequently advancing to liver cirrhosis and hepatocellular carcinoma if not managed with precise prognostic interventions. The capacity to accurately predict patient survival is essential for optimizing resource allocation and treatment planning. Although Machine Learning (ML) has shown promise in medical diagnostics, standard algorithms often underperform when applied to hepatitis datasets characterized by severe class imbalance and high dimensionality. Conventional models tend to bias predictions toward the majority class (survival), resulting in a high rate of False Negatives for the minority class (mortality), which is clinically unacceptable. Moreover, single-classifier approaches often lack the generalization capability necessary for robust clinical deployment. To address these deficiencies, this study proposes a Hybrid Cost-Sensitive Stacking Ensemble Model (HCS-SEM). The framework integrates three strategic components: (1) a rigorous Split-First Synthetic Minority Oversampling Technique (SMOTE) protocol to resolve class skewness without data leakage; (2) a Chi-Square feature ranking mechanism to eliminate redundant clinical attributes; and (3) a Two-Tier Stacking Architecture employing Random Forest, SVM, and Gradient Boosting as base learners, optimized by a Logistic Regression meta-learner. Experimental validation on the UCI Hepatitis dataset demonstrates that HCS-SEM significantly outperforms standalone classifiers and traditional ensemble methods. The model achieves superior performance metrics, particularly in Sensitivity and F1-Score, confirmed by the Friedman Rank Test and Nemenyi post-hoc analysis. These findings suggest that the proposed HCS-SEM provides a robust, clinically viable tool for hepatitis prognosis, offering high-precision decision support for medical practitioners managing high-risk patients.
20 November 2026 in Bulletin of Electrical Engineering and Informatics
Integrating user behavioral factors into non-functional requirements (NFRs) analysis remains difficult because NFRs are often abstract and lack explicit links to human factors. This study presents a cross-domain ontology that semantically maps user behavior dimensions to NFR quality attributes, enabling machine-readable cross-domain reasoning for early-stage NFR elicitation and user-aware requirement prioritization. Following the Methontology framework, knowledge was acquired from two sources: a behavioral dataset from a survey of 267 industrial-sector employees in Indonesia (9 dimensions; 36 indicators; validated via partial least squares structural equation modeling (PLS-SEM), average variance extracted (AVE): 0.746–0.807, CR: 0.898–0.926) and an ISO/IEC 25023-derived NFR indicator set (45 indicators across usability, performance efficiency, and security). Semantic similarity was computed using sentence-BERT (SBERT), producing 1,620 pairwise similarity scores (range: 0.20–0.85) instantiated as first-class individuals in a web ontology language-description logic (OWL-DL) ontology. Validation included reasoner-based consistency checking (Pellet: 2,490 ms), SPARQL competency question testing, and expert agreement evaluation by five domain experts (mean satisfaction: 80%; Cohen's Kappa: 0.37–0.58). The resulting ontology provides a reusable semantic backbone for user-aware NFR analysis and supports future integration with embedding-based semantic matching and intelligent requirement prioritization tools.