Social networks have revolutionized communication, the dissemination of information, and the creation of online communities. However, it has also encouraged the growth of problems such as cyberbullying and, in particular, cyberstalking, a persistent form of online harassment that affects the safety and well-being of victims, especially women, adolescents, and children. Detecting and preventing this phenomenon represents a significant challenge due to the complexity of language, the variability of harassment patterns, and the lack of balanced and appropriately labeled data sets. This study systematically reviews the literature using the PRISMA methodology to identify and analyze Artificial Intelligence (AI)-based models designed to address this problem. Thirty-three primary studies employing machine learning (ML) and natural language processing (NLP) techniques were selected to evaluate their effectiveness in detecting abusive behavior. The findings reveal that algorithms such as support vector machines, convolutional neural networks, and long short-term memory have demonstrated high levels of accuracy in identifying cyberstalking patterns. Furthermore, combining advanced models such as BERT and hybrid strategies that involve ML and Deep Learning (DL) has significantly improved the performance of detection systems. Finally, it is highlighted that the quality and diversity of data sets play a crucial role in optimizing these models, making it necessary to develop more representative datasets and strategies that enable real-time detection to mitigate the impact of cyber-stalking on digital platforms.