Kosimov Abdunabi Abduraufovich,  Asadullozoda Shahriyor Mirzokhon, Sharifov Muso Umarovich

Annotation:

This article investigates the problem of automatic programming language identification based on the statistical analysis of Latin characters and the application of artificial intelligence methods. Given the rapid growth in the volume of source code, the need to develop intelligent methods for classifying programming texts in the fields of information security and digital forensics has become particularly relevant. The study employs methods of statistical text analysis, probabilistic metrics (TF-IDF, bigrams), and modern machine learning algorithms, including neural networks and transformer architectures. The source code is considered as a multidimensional object with a set of unique statistical features. The findings reveal that the distribution of Latin characters, keywords, and syntactic structures is unique to each programming language, acting as a "statistical fingerprint" of the language. Based on a comparative analysis, it was determined that hybrid models combining statistical analysis with neural networks provide the highest level of accuracy (up to 98.1%). The practical significance of the research lies in the fact that the obtained results can be successfully applied in malware detection systems, plagiarism detection tools, intelligent compilers, and big data analysis platforms.

Keywords: artificial intelligence, machine learning, source code, pattern recognition, statistical analysis, neural networks, language, data classification, digital forensics.