Journal of Multimedia Processing and Technologies Sameh Azouzi, Zaki Brahmi, Sonia Ghannouchi 16 2 2025 https://doi.org/10.6025/jmpt/2025/16/2/57-82 https://www.dline.info/jmpt/fulltext/v16n2/jmptv16n2_2.pdf With the proliferation of technology, the field of adaptive e-learning has garnered significant attention in recent years. This is because it has allowed users to learn at their own pace and to define personal learning paths based on their individual interests and needs. Using several different devices and sensors around the world can provide to generate massive amounts of data. The analysis of this collected data will provide a basic solid information to ensure adaptive e-learning. Machine learning and data analytics are today very common techniques that can help extract information and find valuable patterns within the collected data. In this work, the field of adaptive e-learning is investigated in terms of definitions and characteristics. Moreover, a taxonomy of various challenges, used machine learning algorithms, the data used in this process are discussed. Also, some of the works proposed in the literature, which tackle these challenges are presented. Our study shows that, despite attempts made by these works to improve the adaptive e-learning. Data processing is generally performed in deferred time, which does not reflect the current state and needs of learners. Likewise, the learner’s behavior is often unpredictable, it can be influenced by several mental and environmental factors and it changes rapidly over time. Data stream mining is very important in adaptive e-learning which originated many main research directions for this area that merit further exploration and investigation.