Reliability of energy storage systems, for stationary as well as mobile applications, is crucial for their stable long term operation. Among all the components that are susceptible to failure, modules made up of individual batteries determine the useful life of such a system. Hence, estimating health quotients of battery packs in typical energy storage systems takes on a high priority. The computational complexity and large volumes of data required in such calculations are well documented. In this article we present an approach for trend prediction of capacity fade while reducing the amount of test data required. This is accomplished through the use of clustering techniques and a supervised learning system, further reducing computation with the use of a recurrent neural network based system. Data for training and validation, mimicking drive cycle data with multiple current pulses, is provided by extensive charge and discharge experimentation in the lab on a commercially available battery pack.