MACHINE LEARNING MODELING THE IAEA DATABASE

D. M. MITNIK; F. BIVORT HAIEK; A. M. P. MENDEZ; C. C MONTANARI

Congreso; International Symposium on Swift Heavy Ions in Matter (SHIM) & International Conference on Atomic Collisions in Solids (ICACS); 2022

University of Helsinki, Finland

The IAEA database was build gathering published articles by diverse authors. Hence, it isnot standarised in its original form and in many cases different experiments are not consistentto one another. The data may have been presented in distinct units and formats. We devoteda significant effort in the reorganization of the database, unifying units, and arranging the datain a standard (csv) format, enabling the obtention of the information to be easily and quicklyaccesible. Considering that the database contains several dozen of thousands input values,purging manually this data requires also a considerable amount of work. To this purpose,we developed a machine learning method to clean up the database, based on an unsupervisedclustering technique (DBSCAN) which identifies outlier values, and determines which data tokeep in cases of overlapping, taking into account the year of production. The refined data isused to train a deep neural network, able to accurately reproduce the input experimental results,and to predict new results in cases which have never been seen in the training procedure (thetest set). The code ESPNN (Electronic Stopping Power with Neural Network), performing allthese mentioned tasks, is presented in this work.