Objective: The rapid expansion of cloud computing services has significantly increased exposure to advanced cyber threats. This study aims to conduct a comparative systematic review of Intrusion Detection Systems (IDS) based on Machine Learning (ML) and Deep Learning (DL), proposed over the past five years, to identify promising approaches applicable to cloud environments. Methods: A systematic review was conducted following the PICOC and PRISMA-A methodologies. Structured searches were performed in indexed databases (Scopus, ScienceDirect, IEEE Xplore), selecting studies published between 2020 and 2025 in Q1 open-access journals. Inclusion criteria prioritized articles proposing IDS models specifically applied to cloud environments. Results: A total of 34 unique studies were identified that propose IDS models for cloud environments. The findings reveal a predominance of hybrid approaches combining ML, DL, and metaheuristic techniques, many of which report high accuracy (above 95% in several cases). Distributed and lightweight architectures are emerging as viable alternatives for IoT and edge-cloud scenarios. Limitations: Comparability across studies is hindered by heterogeneous datasets, validations conducted in simulated environments, and the lack of consistent evaluations in real-world production settings. Conclusions: Intelligent detection approaches show strong potential to enhance cloud security. However, their performance must be validated in real-world contexts, with further efforts needed to address the detection of evasive attacks and to optimize operational scalability.