Estimating battery state of health (SOH) is important for the reliable operation of battery management systems, especially in hybrid vehicles using nickel–metal hydride batteries. In this study, a data-driven method is presented for estimating the SOH indicator of a NiMH battery module based on voltage and current data. From a manufacturing perspective, the proposed method can help improve quality control, reduce end-of-line testing time, and decrease dependence on long capacity tests. The data were collected using an experimental instrumentation platform including voltage measurement, current measurement, and load control. The raw data were filtered using physical limits to remove abnormal records while preserving both low-current and higher-current operating regimes. The filtered data were then divided into fixed-length windows, and voltage- and current-related features were extracted. Two models, Random Forest and GRU, were evaluated for SOH indicator estimation. The best GRU performance was obtained at a sequence length of 20, with RMSE = 0.005022, MAE = 0.003012, and R² = 0.996177. Under the same comparison condition, Random Forest achieved RMSE = 0.016402, MAE = 0.003950, and R² = 0.959225. The results showed that both models could estimate the SOH indicator, but GRU provided better performance by using temporal information from consecutive windows. The results also showed that excessively increasing the sequence length did not necessarily improve accuracy and could reduce the performance of the recurrent model.