Ensemble Learning Using KNN and Decision Tree for Virus Infection Classification in Mouse Study Dataset

Abstract

In this study, we propose an ensemble learning approach to classify viral infection presence in mice using the Mouse Viral Infection Study Dataset. The dataset includes two numerical features—volumes of two administered medications—and a binary label indicating viral presence. To improve prediction performance, we combined K-Nearest Neighbor (KNN) and Decision Tree (DT) classifiers within a soft voting ensemble framework. Standardization was applied as a preprocessing step to ensure fair feature contribution, especially for the distance-sensitive KNN. The ensemble model underwent hyperparameter optimization using GridSearchCV with 5-fold cross-validation to fine-tune the number of neighbors for KNN and depth-related parameters for DT. The experimental results demonstrated that the ensemble classifier achieved perfect performance, with 100% accuracy, precision, recall, and F1-score on the test set. The confusion matrix showed no misclassifications, and the Receiver Operating Characteristic (ROC) curve achieved an Area Under Curve (AUC) of 1.00, indicating excellent separability between classes. These results suggest that the proposed ensemble effectively leverages the strengths of both KNN and DT, making it suitable for biomedical classification tasks where interpretability and reliability are critical. Although the model performed exceptionally well, the simplicity of the dataset, including balanced classes and clear feature boundaries, may have contributed to the ideal performance. Thus, while the findings are promising, further validation is necessary using more complex or noisy datasets. This study contributes a practical, interpretable, and effective ensemble learning framework for binary classification problems in experimental virology, and opens pathways for further research in preclinical biomedical data analytics using hybrid classification systems.