Machine learning methods have become prevalent in the field of biomedicine due to the development of genetic, imaging, and clinical data. In particular, classification methods are used to predict phenotypes such as patient survival or response to treatment. However, traditional machine learning methods often lack interpretability due to their black-box nature.
This thesis focuses on an interpretable classification method, the basicEnBMC, which constructs an ensemble classifier based on bivariate monotonic classifiers (BMC), to predict binary outcomes. We propose two extensions: (1) we reduce the time complexity of the basicEnBMC approach by introducing a preselection step, leading to the fastBMC method, and (2) we extend our classifiers to ordinal and multiclass classification problems, generalizing the preselection and developing the fastMBMC method.
These models were successfully applied to various transcriptomic datasets and compared with other ordinal classification models. Our methods have effectively identified biological mechanisms leading to phenotypes, such as predicting the recurrence time in breast cancer. They have enabled the identification of known biomarkers and pathways, indicating future research potential. In the field of food science, we demonstrated the broader applicability of these models.