Feature Relevance Bounds for Ordinal Regression
This work addresses the need for better interpretability in ordinal regression models, particularly for sociodemographic data, but appears incremental as it builds on existing approaches to handle variable dependencies.
The paper tackles the problem of interpreting ordinal regression models in high-dimensional or correlated data by proposing feature relevance bounds to identify and differentiate between strongly and weakly relevant features, aiming to improve model interpretability beyond existing sparse methods.
The increasing occurrence of ordinal data, mainly sociodemographic, led to a renewed research interest in ordinal regression, i.e. the prediction of ordered classes. Besides model accuracy, the interpretation of these models itself is of high relevance, and existing approaches therefore enforce e.g. model sparsity. For high dimensional or highly correlated data, however, this might be misleading due to strong variable dependencies. In this contribution, we aim for an identification of feature relevance bounds which - besides identifying all relevant features - explicitly differentiates between strongly and weakly relevant features.