Decomposition and Combination-based Classifier Chains for Semi-Supervised Multi-Dimensional Classification
Zelin Xu, Peipei Li · 2023
In multi-dimensional classification (MDC), each training example is represented by a single instance which is associated with multiple class variables from different class spaces. However, in real-world applications, only a limited size of labeled data is available to learners due to the expensive annotation cost. Therefore, this paper proposes the decomposition and combination-based classifier chains (DCCC) to solve the semi-supervised multi-dimensional classification problem (SSMDC). Specifically, DCCC firstly builds the binary classifier chains via the one-vs-one decomposition strategy to consider dependencies among class spaces. Secondly, in the training and prediction process of the chain, the final prediction results are obtained by combining the predictions of the model trained solely on labeled samples and the predictions obtained through label information communication. Then, considering the performance of DCCC will be affected by a specific order, an integrated version of the above approach is proposed. Finally, comparative experiments with one SSMDC approach and the state-of-the-art MDC approaches clearly verify the effectiveness of the proposed approach.