Ordinal Classification Using Single-Model Evidential Extreme Learning Machine

The extreme learning machine model for ordinal classification is extended to the uncertain case. Dealing with epistemic uncertainty by Dempster-Shafer theory, in this paper, the single-model multi-output extreme learning machine is learned from evidential training data. Taking both the uncertainty and the ordering relation of labels into consideration, given mass functions of training labels, different evidential encoding schemes for model output are proposed. On that basis, adopting the structure of a single extreme learning machine model with multiple output nodes, the construction procedure of evidential ordinal classification model is designed. According to the encoding mechanism and learning details, when there is no epistemic uncertainty in training labels, the proposed evidential ordinal method can be reduced to the traditional ordinal one. Experiments on artificial and UCI datasets illustrate the practical implementation and effectiveness of proposed evidential extreme learning machine for ordinal classification.