Intelligent Cybersecurity: Applying Machine Learning Techniques for the Detection and Prevention of Digital Threats

Abstract

Digital transformation has increased the exposure of systems, networks, devices, applications and data to threats capable of altering their characteristics and making them more difficult to identify through conventional mechanisms. In this setting, machine learning techniques have gained relevance because of their ability to process information, recognise patterns and distinguish behaviours associated with potentially malicious activities. The study aimed to examine specialised studies addressing the use of these techniques for the detection and prevention of digital threats, considering the algorithms employed, threats examined, detection capabilities, performance criteria, contributions and limitations. A qualitative, descriptive and documentary approach was adopted, using PRISMA 2020 to structure the identification, selection and screening of sources. The findings show applications of supervised algorithms, unsupervised methods and deep learning architectures in the detection of intrusions, malware and anomalies. Supervised models facilitate the classification of previously characterised threats, whereas unsupervised procedures enable atypical behaviours to be identified. Deep learning architectures extend the processing of complex, high-dimensional information. Limitations persist in relation to data quality and representativeness, class imbalance, false positives, interpretability and the continuous evolution of digital threats, all of which constrain the capabilities attributed to these techniques.