A Survey of Interpretable Machine Learning Methods

Recently, people are increasingly pursuing more accurate machine learning systems. At the same time, the problem is that it is difficult for people to understand the internal logical structure of the system, and the machine learning model has become a black box. Lack of explanation is not only a practical problem, but also an ethical problem. At this point, understanding why a model makes a prediction is as important as the accuracy of the prediction. In order to balance the interpretability and accuracy of the model, many experts and scholars have done a lot of research on improving the interpretability of machine learning model, improved the interpretability of the model from different dimensions, and created a series of interpretable methods to help users understand the internal logic of the system and make the black box transparent. However, the current interpretability research is still in the stage of development, and there is no completely unified interpretable method at present. In this paper, firstly, the development status of machine learning and the importance of interpretable learning are discussed. Then, according to the current research status, interpretable methods are introduced from interpretable models and model independent categories, the research and application in the field of interpretability are discussed, and the current challenges and future development of interpretability are analyzed, To further promote the research and development of interpretability.