The problem of task scheduling in distributed environments such as Cloud Computing is seen as a major challenge for ensuring efficient resource management and improving overall system performance. Traditional approaches to task scheduling based on heuristics and meta-heuristics have proven useful in some scenarios, but they have limitations. Given that Cloud Computing is a complex and dynamic environment, the challenge is to create a scheduling strategy that has the ability to adaptively understand the nature of this environment, and machine learning (ML) techniques are emerging as promising tools for this. To focus on this direction, we provide a review of ML-based approaches in a very specific context: dynamic scheduling of independent tasks in Cloud Computing. Through a deep analysis of these approaches, we also propose a comparative study between these strategies in terms of optimized metrics. We also underline the growing interest in hybrid approaches that combine ML techniques with meta-heuristics.