Nowadays, more and more emphasis is given to automatic systems that can make decisions without human intervention. These decisions can cover huge areas of life, and are found in many industries, like finance, economics, manufacturing or healthcare.
Problems which have a huge economic or moral impact, there is a need to gain insights into how our models work. On the one hand, we can examine and acknowledge the correctness of models, on the other we can comprehend models that are difficult to understand in terms of their internal functioning and decision-making process.
In my thesis, I will develop a system that helps to gain insights into different models by representing the dependence between individual input variables and outputs in an easily understandable form to the user. In this way, we can better understand the correlation of the variables of more complex models (such as random forest) with the output, compare it with simpler models (linear or logistic regression), and use these models in production environments.
In the dissertation, I will give a detailed description of the task and the presentation of the technologies used. I describe the implementation details on both the server and the client side. I will compare several independent data sets and present the partial dependencies discovered by the system. Finally, I will explore some of the possibilities for further development of the system.