1 January 2026 in Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science
Accurate and scalable identification of microorganisms remains a fundamental challenge in biomedical diagnostics and environmental monitoring due to the trade-off between accuracy, interpretability, and computational efficiency in existing approaches. While deep learning models achieve high performance, they often require large annotated datasets and lack transparency, limiting their practical adoption. This study proposes an interpretable and computationally efficient machine learning framework for microorganism classification based on structured morphometric features. The approach is evaluated on the Microbes Dataset, comprising 30,527 samples with 23 quantitative descriptors across 10 taxonomic classes. A unified pipeline integrating preprocessing, feature scaling, and hyperparameter optimization via GridSearchCV with stratified cross-validation is developed. The performance of K-Nearest Neighbors, Support Vector Machine, and Random Forest is systematically compared. The results show that Random Forest achieves superior performance, reaching 99.00% accuracy and an F1-macro score of approximately 98.8%, while maintaining high interpretability. Feature importance analysis reveals that geometric descriptors dominate classification decisions, and hierarchical clustering independently confirms the intrinsic structure of the data. The key contribution is demonstrating that ensemble learning on morphometric features can match deep learning performance while ensuring interpretability, data efficiency, and suitability for real-time deployment in practical microbiological systems.