1 September 2025 in BAREKENG: Jurnal Ilmu Matematika dan Terapan
Diabetes mellitus is a chronic condition with high blood sugar that can cause severe organ damage, affecting all ages globally. Early diagnosis is crucial for improving patients' quality of life, and machine learning offers a promising approach. The Support Vector Machine (SVM) is effective for classification, but feature selection is essential to enhance the relevance of features. The Whale Optimization Algorithm (WOA) is an optimal method for global feature selection, but it has a drawback-premature convergence, which can lead to suboptimal results. This issue should be addressed by modifying mutation operations, convergence factors, and population initialization, resulting in Binary Gaussian IWOA (BGIWOA). This research focuses on feature selection using BGIWOA, comparing it with Variance Inflation Factor (VIF) using SVM. The result show that BGIWOA is better than VIF and the best configuration BGIWOA’s parameter is with linear kernel. This configuration produces the best accuracy of 95.00%. BGIWOA-SVM demonstrates better accuracy with stable consistency compared to VIF-SVM. The best SVM model achieves average accuracy of 95.62% for training data and 95.58% for validation data, with an accuracy of 93.85% for the test data. This model also yields an average precision of 94.00%, a recall of 91.00%, and an -score of 92.00%. The model was also better than SVM without optimization, which only achieved a training accuracy of 84.25% and a testing accuracy of 81.30%. This model can assist in diagnosing diabetes with accurate and consistent predictions for new data. The results are specific to the diabetes dataset used in this research, so further testing on other binary datasets is necessary to confirm the model's effectiveness and generalizability across different domains and types of data.
15 June 2025 in CAUCHY: Jurnal Matematika Murni dan Aplikasi
IMI Gusti Ngurah Bagus Ferry Mahayudha
I Gusti Ngurah Bagus Ferry Mahayudha
Brain tumor detection remains a significant challenge due to the complex variations in tumor appearance. Although deep learning models have demonstrated high accuracy, their limited interpretability hinders clinical adoption. To address this issue, this study integrates Gradient-weighted Class Activation Mapping (Grad-CAM) into Convolutional Neural Networks (CNNs) to enhance the visual interpretability of predictions. Grad-CAM extends Class Activation Mapping (CAM) and is applicable to a wide range of deep learning architectures. The primary contribution of this work is the demonstration that combining Grad-CAM with MobileNet architectures yields an interpretable and efficient framework for diagnosis of brain tumor, effectively balancing accuracy, computational efficiency, and clinical transparency. Using a Brain Tumor MRI dataset, MobileNetV4 achieved an accuracy of 98.29% with the shortest training time (1738.82 seconds) and an ROC accuracy of 99.96%. MobileNetV3 achieved 99.62% accuracy with an ROC accuracy of 99.92%. Grad-CAM effectively highlighted tumor regions while showing uniform attention in non-tumor cases, thereby reducing false positives. These results demonstrate that lightweight models can achieve a strong balance between predictive performance, training efficiency, and interpretability. The proposed framework thus supports the development of explainable and efficient diagnostic tools for clinical practice.
1 January 2026 in Applied Computational Intelligence and Soft Computing
Accurate feature selection remains a critical challenge in heart disease prediction due to high‐dimensional data and complex feature interactions. This study proposes a hybrid framework that integrates extreme gradient boosting with a chaotic binary bat algorithm to improve feature‐selection efficiency and predictive performance. The proposed approach incorporates chaotic dynamics to enhance search stability and reduce premature convergence in binary optimization. Experimental results on two benchmark cardiovascular datasets demonstrate that the proposed model consistently outperforms conventional feature‐selection methods and metaheuristic algorithms, including genetic algorithm and particle swarm optimization. On Dataset I, the model achieves a test accuracy of 0.917 ± 0.005 and an F1‐score of 0.926 ± 0.004, representing an improvement of approximately 3.3%–3.4% over baseline extreme gradient boosting. On Dataset II, both the proposed method and the binary bat algorithm achieve identical optimal performance (F1‐score = 1.000), indicating high dataset separability. Furthermore, the proposed approach reduces computational time by up to 2.5 times compared to the genetic algorithm and particle swarm optimization. The findings indicate that feature selection is the primary factor driving performance improvement, while chaotic dynamics mainly enhance convergence efficiency and stability. This study contributes a robust and efficient feature‐selection framework that effectively balances accuracy, computational cost, and interpretability for heart disease prediction.
17 October 2026 in CAUCHY: Jurnal Matematika Murni dan Aplikasi
This study proposes a transaction cost-aware portfolio optimization framework for NASDAQ-100 stocks based on the Hippopotamus Optimization Algorithm (HOA). The mathematical model extends the classical mean-variance framework by incorporating transaction costs into a Net Sharpe Ratio (NSR) objective function. Daily adjusted closing prices of NASDAQ-100 constituent stocks from 2021 to 2025 are employed, with the twenty highest-ranked stocks according to the Sharpe Ratio (SR) selected as the candidate investment universe. Each optimization experiment is independently repeated 25 times to evaluate robustness and solution stability. The results indicate that the proposed HOA consistently achieves competitive optimization performance with very small variability across repeated runs throughout the investment horizon. Although several competing algorithms attain comparable or slightly better objective values during particular rebalancing periods, HOA remains among the strongest-performing methods while exhibiting stable convergence behavior. Throughout the investment horizon, the proposed algorithm produces the highest average cumulative portfolio value of \$5,307.87 \$125.71 from an initial investment of \$1,000, corresponding to an average annual return of 40.03% 0.68%, together with the highest average SR and NSR of 1.5736 0.0151, while maintaining competitive maximum drawdown and transaction costs. These findings demonstrate that incorporating transaction costs directly into the optimization objective enables HOA to achieve a favorable balance between portfolio growth, risk-adjusted performance, and trading efficiency under dynamic market conditions.