For the development, simulation and validation of data-driven battery aging models, a critical aspect is having access to large amounts of reliable aging data. Although normal operation of battery packs can be simulated in the lab to generate aging data, a variety of other non-operational profiles are typically needed, requiring many hours of testing, often at conditions different than normal operational conditions observed when the battery pack is deployed in its intended application. Moreover, application of prolonged and multiple capacity tests can be detrimental to the health of the battery. In view of these concerns, this article continues a line of research into capacity fade estimation approaches that require less data and time for the data generation process; in particular, an approach using rule based machine learning for Li-ion battery packs is proposed. Using data generated in the laboratory, aging behavior is characterized by measurable features and a supervised learning approach in order to estimate capacity fade using real-time operational data toward the goal of eliminating the need for specific capacity tests. The experimental results presented in this article focus on proof of concept and are part of a comprehensive study into general capacity estimation and capacity fade estimation in battery packs.