Techniques for computing a memory health metric are disclosed. A system computes a memory health metric as a function of estimated entropy in a data set. The system may determine the memory health metric based on compressed and uncompressed sizes of the data set, quantities of values in the data set, and/or patterns of structural relationships within the data set. The system determines whether the memory health metric crosses a threshold value. If the memory health metric crosses the threshold value, the system presents a warning indicating that utilization of the memory is unhealthy.