Automatic Security Classification with Lasso

With an increasing amount of generated information, also within security domains, there is a growing need for tools that can assist with automatic security classification. The state-of-the art today is the use of simple classification lists "dirty word lists" for reactive content checking. In the future, however, we expect there will be both proactive tools for security classification assisting humans when creating the information object and reactive tools i.e. double-checking the content in a guard. This paper demonstrates the use of machine learning with Lasso Least Absolute Shrinkage and Selection Operator [1, 2] both to two-class binary and multi-class security classification. We also explore the ability of Lasso to create sparse solutions that are easy for humans to analyze and interpret, in contrast to many other machine learning techniques that do not possess an explanatory nature.