Abstract
ABSTRACT The automation of information security auditing requires effective software analysis methods, particularly when executable file metadata is missing, corrupted, or intentionally modified. This study presents a comparative analysis of machine learning methods for identifying processor instruction architectures using byte‐frequency characteristics of software. Averaged byte‐frequency signatures from different architectures were used to evaluate Naïve Bayes models, metric‐based methods, linear classifiers, decision trees, random forests, support vector machines, neural networks, and ensemble models. Classification performance was assessed using Accuracy, Precision, Recall, and F1‐score metrics. The results demonstrate that machine learning techniques provide satisfactory classification performance even with limited training data and significantly outperform random classification. Among the evaluated approaches, the Linear SVC (Support Vector Classifier) achieved the highest performance across all metrics. Error analysis revealed groups of closely related architectures that differ in endianness, bitness, and floating‐point implementation, which represent the primary source of misclassification. The findings confirm the potential of machine learning methods for automating information security audits, software integrity verification, and digital forensic investigations.