Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2023
- Goals
- Record
- b821c3c7-0a65-4638-8c25-2b75eeaf769c
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a health research activity on the multi-class classification of dengue infection cases using imbalanced clinical diagnosis data. The activity addressed the problem that imbalance in a dengue infection dataset can affect the interpretation of predicted results because classification of the minority class may be biased toward the majority class. The activity was conducted by researchers Fahmi, Amiq; Muqtadiroh, Feby Artwodini; Purwitasari, Diana; Sumpeno, Surya; and Purnomo, Mauridhi Hery.
The activity aimed to improve classification accuracy by addressing multi-class imbalance through a proposed approach that assigned class weights to minority and majority classes. Resampling of imbalanced datasets was also conducted using Random resampling and SMOTE techniques. The balanced datasets were evaluated with eight classification algorithms: NN, KNN, Decision Tree, Random Forest, Naive Bayes, AdaBoost, SVM, and Logistic Regression. The evaluation applied 10-fold cross-validation and feature selection.
The experimental results showed that the proposed approach improved accuracy compared with the original primary data. Among the tested algorithms for dengue infection cases, AdaBoost produced the highest accuracy, reported as 87.0%. The activity also tested the new method on hypothyroid disease cases to demonstrate its effectiveness and efficiency in increasing accuracy. In those cases, AdaBoost produced the highest reported accuracy of 99.7%, with average AUC, F1, precision, and recall toward 99.8%.
Through this activity, UDINUS addressed a health-related research problem involving dengue infection classification and imbalanced datasets. The activity contributes to the stated goal through the use of classification, resampling, class weighting, cross-validation, and feature selection to examine accuracy in clinical diagnosis data. The reported findings frame the proposed method as applicable to classification problems in imbalanced datasets.