Hybrid Classification by Integrating Expert Knowledge and Data: Literature Review

At present, the common classification methods can be divided as knowledge-driven and data-driven. The knowledge-driven classification methods have high performance in terms of interpretability, but they fail to consider the distribution within the data and can not take the relationship between the data into consideration, which lead to poor classification accuracy. By contrast, the data-driven classification methods have excellent performance in classification accuracy, but poor performance in interpretability. Combining these two types of classification methods to build a hybrid classification model, is probable to achieve a compromise between accuracy and interpretability, which is of great importance for those applications where both the classification accuracy and model interpretability are needed, e.g., medical diagnosis. Therefore, this paper mainly reviews the current research on hybrid classification methods.