Predicting nuclear fuel parameters by using machine learning techniques

Today, machine learning techniques are widely used to solve complex problems at computing level. One of these techniques is the vector support machine which has been used in various applications. This technique will be used to solve a problem relating to the nuclear industry and its control through the Treaty on the Non-Proliferation of Nuclear Weapons (NPT). Specifically, this work seeks to create a model capable of predicting certain parameters of importance in irradiated fuels, under an analysis of a number of characteristics. For this purpose, a simulated database, with a significant number of irradiated fuels, was used. The analysis process began with a treatment of the information, with the purpose of using machine learning tools. Subsequently, a validation process of the built model took place, where experimentation will contrast two models with different amounts of fuel characteristics. Finally, an Boruta analysis is carried out to analyze and obtain a measure of the importance over the importance parameters. The results showed that the two models obtained a high perform even when the number of features were substantially different. Even models with the use of hyperparameter tuning improve performance, as demonstrated below. (author)