Battery health evaluation and management are vital for the long-term reliability and optimal performance of lithium-ion batteries in electric vehicles. Electrochemical impedance spectroscopy (EIS) offers valuable insights into battery degradation analysis and modeling. However, previous studies have not adequately addressed the impedance uncertainties, particularly during battery operating conditions, which can substantially impact the robustness and accuracy of state of health (SOH) estimation. Motivated by this, this paper proposes a comprehensive feature optimization scheme that integrates impedance validity assessment with correlation analysis. By utilizing metrics such as impedance residuals and correlation coefficients, the proposed method effectively filters out invalid and insignificant impedance data, thereby enhancing the reliability of the input features. Subsequently, the extreme gradient boosting (XGBoost) modeling framework is constructed for estimating the battery degradation trajectories. The XGBoost model incorporates a diverse range of hyperparameters, optimized by a genetic algorithm to improve its adaptability and generalization performance. Experimental validation confirms the effectiveness of the proposed feature optimization scheme, demonstrating the superior estimation performance of the proposed method in comparison with four baseline techniques.