Software defects, errors, or weaknesses in the software's implementation, design, or architecture can all result in vulnerabilities. There is a potential to discover security vulnerabilities and learn about bug patterns as a result of the expansion of open-source code that is now available for analysis. The explosion has given rise to an opportunity. Recent developments in deep learning for image processing, audio identification, and natural language processing have shown how powerful neural models can be at understanding natural language. Many individuals are currently interested in the research of vulnerability identification, and over the past few decades, specialists have offered a wide range of methodologies. The detection of malware, the identification of software vulnerabilities, and the recognition of functions are just a few of the numerous real-world applications in which machine learning (ML) has been successfully used as a method for learning high-quality feature representations. Other examples include those previously mentioned. This is a comprehensive summary of the research that has been conducted in the field of detecting software vulnerabilities using machine learning techniques. First, we'll go over the fundamental ideas that underpin each of the three primary approaches to discovering vulnerabilities: static analysis, symbolic execution, and fuzzing. These detection methods are extremely common and frequently employ more standard methods. They do not include the most popular machine learning methods, such as supervised learning and deep learning techniques. When we are better prepared, we will be able to anticipate and focus on the research challenges in software vulnerability detection that require immediate attention. There are various possible uses for machine learning techniques in software measurement. Software defect prediction and software work estimation are two of the machine learning techniques most frequently used in software maintenance. The artificial neural network for SM is the most commonly used machine learning technology today. NASA and Promise databases are frequently used in empirical research. Over the last decade, the SM paradigm has gradually shifted away from ensembles of single ML models and toward deep learning models. Several factors contributed to this shift. Machine learning (ML) methodologies have typically produced favourable results when applied to a wide range of different prediction endeavours.

Software defects, errors, or weaknesses in the software's implementation, design, or architecture can all result in vulnerabilities.