Caches are an important component of modern computing systems given their significant impact on performance. In particular, caches play a key role in the cloud due to the nature of large-scale, data-intensive processing. One of the key challenges for the cloud providers is how to share the caching capacity among tenants, while each often requires a different degree of quality of service (QoS). The invariant is that the individual tenants' QoS requirements should be preserved while the cache usage is optimized in a system-wide manner. To this end, this research proposed a learning-based approach for dynamic cache management in the cloud through the prediction of the cache performance based on the tenants' data access patterns. This work modeled the data access patterns using a diverse range of the probabilistic distributions and established a set of learning-based regression models to continuously predict the cache hit rate based on the estimated distribution model. The predicted performance is in turn referenced to dynamically reallocate the cache space for the tenants in real time. Experimental results of this thesis with a set of synthetic traces and the Yahoo! cloud serving benchmark show that the proposed method consistently optimizes the cache space while preserving the QoS requirements.