Interpretability of machine learning models

As machine learning algorithms have been used in many areas of our lives, for example, self-driving cars, healthcare, and the financial industry, the problem of trust is becoming even more urgent. To trust the decisions that the algorithms adopt, we need to understand the nature of their occurrence, so often we need not only a theoretical understanding of their work, but also special tools that would explain the origin of the findings within the algorithms themselves and present the withdrawn information in an informative and accessible form. This article will list some problems related to the interpretation of machine learning algorithms, as well as the desired properties of interpreted models that can improve the perception of algorithms and increase people's confidence in the decisions made. In the following, some visual analytics tools will be discussed, as well as one of the model-agnostic methods, LIME, which studies the model locally around the prediction and explains any classifier.