Individual cells within isogenic tumor populations can exhibit distinct cellular morphologies, behaviors, and molecular profiles. Cell state plasticity refers to the propensity of a cell to transition between these different morphologies and behaviors. Elevation of cell state plasticity is thought to contribute to critical stages in tumor evolution, including metastatic dissemination and acquisition of therapeutic resistance. However, methods for quantifying general plasticity in mammalian cells remain limited. Working with a HoloMonitor M4 digital holographic cytometry platform, we have established a machine learning-based pipeline for high accuracy and label-free classification of adherent cells. We use twenty-six morphological and optical density-derived features for label-free identification of cell state in heterogeneous cultures. The system is housed completely within a mammalian cell incubator, permitting the monitoring of changes in cell state over time. Here we present an application of our approach for studying cell state plasticity. Human melanoma cell lines of known metastatic potential were monitored in standard growth conditions. The rate of feature change was quantified for each individual cell in the populations. We observed that cells of higher metastatic potential exhibited more rapid fluctuation of cell state in homeostatic conditions. The approach we demonstrate will be advantageous for further investigations into the factors that influence cell state plasticity.