Numerous empirical models have been proposed to predict the behavior of reinforced concrete structural walls (RCSWs), often relying on the assumption of linear relationships between variables. However, under seismic conditions, RCSWs tend to exhibit nonlinear and complex behaviors that such traditional models may fail to accurately capture. Consequently, there is increasing interest in applying machine learning (ML) methods, which are particularly effective at modeling nonlinear relationships and capturing the complexities inherent in seismic performance. This study applies ML techniques to analyze the behavior of RCSWs, focusing on predicting drift capacity and failure modes, while also assessing the predictive accuracy and interpretability of the ML-based models. The interpretability of the models provides valuable insights, enhancing the understanding of shear wall behavior and improving the reliability and transparency of ML applications in practical scenarios. At the same time, it is important to consider the trade-off between model accuracy and complexity, as more complex models may offer only marginal improvements in accuracy while increasing computational costs, making them less beneficial in cases where existing models are already sufficiently accurate for decision-making. In light of this, this study presents a reliability-based investigation of the tangible benefits provided by ML models in terms of structural design and performance. To quantify these benefits, the increase in predictive accuracy is interpreted as a reduction in epistemic uncertainty. The case study provides insights into how much improvement in accuracy (i.e., ML relative to traditional models) is needed to have a tangible effect on the seismic design and performance. Furthermore, the study addresses the limitations of the FEMA P-58 framework in assessing seismic damage and economic loss for RCSWs, which currently overlooks failure modes. It develops distinct RCSW fragility functions based on specific failure modes, allowing for more precise damage evaluation and loss assessment. A comparative case study is presented, comparing the results of FEMA P-58 with those of the proposed framework and emphasizing the importance of a probabilistic approach. The findings highlight the value of failure mode-dependent assessment, particularly at lower hazard levels, and demonstrate the benefits of this refined methodology for optimizing structural designs and seismic retrofit strategies.