Reliable estimation of lithium-ion battery State of Health (SoH) during initial phases of operation is crucial to ensure reliability and enable successful use in downstream applications. Conventional offline SoH estimation techniques rely on long-term cycling information and cannot be used for real-time forecasting. Given paper introduces an online early-life SoH prediction model based on the Streaming Random Patches Regressor (SRPRegressor). This incremental ensemble learning algorithm can be updated at the end of every cycle. A 3000 mAh LiFePO4 (LFP) cell was cycled under regulated charge-discharge cycles to produce 180 cycles of working data. Three physically significant features, such as discharge capacity, average discharge voltage, and voltage hysteresis, were obtained per cycle, and actual SoH was calculated based on discharge capacity obtained in each cycle. Data were split into 80% for online training and 20% for online testing, and both phases were implemented in a streaming, cycle-by-cycle manner to simulate live deployment. According to experiment’s results, proposed streaming model yields a correct early-life SoH prediction with a Mean Absolute Error (MAE) of 0.0934. These results show that online ensemble learning is suitable for real-time battery health assessments with minimal input features.