Yeasts Automated Classification with Extremely Randomized Forests

Several pathogenic yeast species are resistant to pharmaceutical agents and have evolved so quickly over time that often they cannot be stopped with antifungal treatments. In particular, Candida species can cause yeast infections in patients’ blood or organs, and Candida outbreaks are particularly troublesome, often found in hospitals and healthcare facilities where people are most susceptible because of their weakened immune systems. Candida can live for weeks on walls and other surfaces, such as walls or furniture, becoming a major hidden danger. New, rapid, and reliable methods of classification and spread prevention are needed. In this preliminary work, we have considered off-the-shelf standard classifiers for yeast classification and applied them to Candida classification. Since performances of classifiers based on logistic regression, SVM, and Extremely Randomized Forests were not particularly remarkable, the possibility to improve their precision has been explored, applying different techniques of feature extraction and feature selection. Eight different Candida strains have been analyzed and systematically tested for classification. The experimental results show that classification with Extremely Random Forests and feature extraction can achieve promising results with the best precision in the automatic classification of candida yeasts, thus being a powerful tool for the prevention and treatment of candida yeasts outbreaks.