Identification of Processor’s Architecture of Executable Code Based on Machine Learning. Part 3. Assessment Quality and Applicability Border.

Abstract

The article presents the author's method testing results for identifying the processor architecture of the executable code based on machine learning. In the third final part of the cycle, its qualitative indicators are determined: accuracy, completeness and F-measure for the executable files of the Debian build. There are investigated the applicability limits of the architecture identification method for four conditions: the file header absence, different sizes of machine code, partial code destruction, and the presence of instructions from several architectures. We can observe the identified disadvantages of the proposed method and ways to eliminate them, as well as the further direction of its development.