Recent advances in machine learning and data analytics combined with the availability of high-performance computational resources have a transformative impact on data-driven modeling. In civil engineering, applications of data-enabled modeling, analysis, and control of complex structural systems can be found across multiple areas such as design optimization, reliability analysis, structural health monitoring, and multi-hazard modeling and prediction.This dissertation focuses on two directions of developing and deploying efficient algorithms for data-driven modeling, identification, and discovery of features embedded in complex engineering systems. The first direction involves leveraging recent developments in deep learning to construct surrogate models to alleviate the computational burden of problems that require repetitive simulations, such as uncertainty quantification and propagation. The emphasis is on addressing two challenges of surrogate modeling: the curse of dimensionality, and model uncertainty. The second direction is focused on extracting interpretable and generalizable patterns of dynamic systems from big spatio-temporal data. This direction is guided by fusing elements from operator-theoretic approaches using recently developed decomposition schemes.