Lenin Rivadeneira Puetate. (2020). Comparative Analysis Of Current Trends For Explainable Artificial Intelligence. Final Career Project (TFM). Universidad Politécnica de Madrid, ETSI Telecomunicación.
The study that was carried out shows some of the various techniques that are currently being used to explain how the different models of Artificial Intelligence (AI) or Machine Learning (ML) behave and how decisions are made, making use of some of these models, and in turn comparing their performance. Today artificial intelligence can be found in many aspects of our daily life. From the simplest aspect, to say an example, the business of selling cosmetics or whether it is in something more relevant such as engineering, medicine, law, electronics, business administration, in short, in all muscles that humans believe is relevant, which is why it will eventually be necessary for AI-powered systems to be sufficiently safe, reliable, accessible and above all to meet the expectations that society demands, and above all to have a relative understanding of why a model does a certain action or makes a certain decision. To do this, in this work first, two data sets have been chosen, from the UCI machine learning repository, which is frequently used by the entire machine learning community, As a second step, once it was chosen with which data sets to work, to treat the data from them, we proceeded to use various AI / ML techniques in such a way that it could be seen how one varied many times with respect to another and others. Sometimes very little between them, obtaining interesting results that are later, in the third part of the work it serves as a starting point to start with the process of explaining each AI model, for this what was done was to apply various techniques to help us with these explanations, first, we will see explanations that by their nature are compatible with each of the models that are implemented, being quite graphic techniques of how an element in general ”the most relevant” has a lot to do with the making of decision of each model, and then other techniques are found separately for certain models with which they maintain better compatibility with other techniques, in the same way, it is explained as the way I made a decision or made a discernment.