Seismic Damage Indices (SDI) are numerical parameters used to estimate the respective damage level of structures. Many formulae have been proposed for their proper definition, based on different assumptions concerning the seismic behavior of structural members after their yielding. The Maximum Interstory Drift Ratio (MIDR) is one of the most widely proposed SDIs. The conventional procedure of its calculation can be achieved by performing Nonlinear Time History Analysis (NTHA) and the appropriate post-processing of its results. This research, employs Machine Learning (ML) algorithms that successfully estimate MIDR of Reinforced Concrete buildings that were designed according to EN1998-1 provisions. Machine Learning modeling can significantly enhance the efficiency of this task, especially in cases where the evaluation of a large number of buildings is required in a limited amount of time. According to the optimal policy, this task should be carried out before a strong seismic event when the application of time consuming NTHA is practically infeasible. The proposed methodology aims to bridge the gap between the well-known capabilities of ML to tackle complex problems, and the need for analytical equations that can be adopted into seismic codes. To this end, a Light Gradient Boosting Machine (LightGBM) model has been trained and it has achieved excellent accuracy. Subsequently, the SHapley Additive exPlanations (SHAP) methodology has been employed to construct analytical equations that approximate the behavior of the underlying LightGBM model. The obtained equations achieved comparable accuracy to the fully trained model. This demonstrates the potential applicability of the proposed methodology to tackle similar complex and impactful engineering challenges.