A novel data pre-processing method is proposed for battery state-of-health (SOH) estimation, aiming to automate the process and support the development of battery digital twins. The method converts one-dimensional (1D) voltage data into two-dimensional (2D) data using a sliding window, enabling automatic feature extraction during machine learning (ML) training. The SOH is estimated by forecasting battery voltage in subsequent cycles. The proposed technique is evaluated on the NASA public dataset for Li-ion battery degradation analysis in four scenarios, showing a significant reduction in root-mean-square error (RMSE). The method eliminates the need for manual feature extraction, enhancing SOH estimation accuracy without increasing model complexity. The approach involves data cleaning, resampling, normalization, and 2D data set creation. A CNN-LSTM model is used for automatic feature extraction and SOH prediction. The performance of the proposed method is validated using MAE and RMSE metrics, demonstrating improved accuracy compared to traditional methods. The results show that the proposed pre-processing technique significantly enhances SOH estimation accuracy, making it a crucial step toward battery digital twin development.A novel data pre-processing method is proposed for battery state-of-health (SOH) estimation, aiming to automate the process and support the development of battery digital twins. The method converts one-dimensional (1D) voltage data into two-dimensional (2D) data using a sliding window, enabling automatic feature extraction during machine learning (ML) training. The SOH is estimated by forecasting battery voltage in subsequent cycles. The proposed technique is evaluated on the NASA public dataset for Li-ion battery degradation analysis in four scenarios, showing a significant reduction in root-mean-square error (RMSE). The method eliminates the need for manual feature extraction, enhancing SOH estimation accuracy without increasing model complexity. The approach involves data cleaning, resampling, normalization, and 2D data set creation. A CNN-LSTM model is used for automatic feature extraction and SOH prediction. The performance of the proposed method is validated using MAE and RMSE metrics, demonstrating improved accuracy compared to traditional methods. The results show that the proposed pre-processing technique significantly enhances SOH estimation accuracy, making it a crucial step toward battery digital twin development.