Cells dynamically change their internal organization via continuous cell state transitions to mediate a plethora of physiological processes. Understanding such continuous processes is severely limited due to a lack of tools to measure the holistic physiological state of single cells undergoing a transition. We combined live-cell imaging and machine learning to quantitatively monitor skeletal muscle precursor cell (myoblast) differentiation during multinucleated muscle fiber formation. Our machine learning model predicted the continuous differentiation state of single primary murine myoblasts over time and revealed that inhibiting ERK1/2 leads to a gradual transition from an undifferentiated to a terminally differentiated state 7.5-14.5 hours post inhibition. Myoblast fusion occurred ~3 hours after predicted terminal differentiation. Moreover, we showed that our model could predict that cells have reached terminal differentiation under conditions where fusion was stalled, demonstrating potential applications in screening. This method can be adapted to other biological processes to reveal connections between the dynamic single-cell state and virtually any other functional readout.