Heterogeneous and multi-structured data, stored at distributed geographical locations leads to latency in user query processing and unavailability of demanded data. The existing caching schemes based on the duration of the web page stay within the cache, lag while dealing with the heterogeneity of content’s demand, and fail to provision dynamic caching automatically. In this context, this paper proposes a novel dynamic and automatic cache management model named QuAd-Caching. It integrates diverse learning with the computational efficiency of Quantum machine learning and optimal solution-finding capability of Adam optimization for proactive estimation of caching contents. Specifically, three distinct QuAd estimators for cache size prediction, eviction, and entry, are employed to capture all-inclusive dynamic cache management in diverse data lake environments. The simulation and performance evaluation of the proposed QuAd caching using a benchmark dataset confirms its efficiency and potency in dynamic cache management while reducing average data access time up to 100.826 nsec as compared with optimal case reporting 100 nsec, and minimizing average delay up to 99 % over without QuAd-caching. Further, the number of cache hits is improved up to 52.7 % over the existing caching approaches.