Ground motion models (GMMs) are developed using the past strong motion records to predict the ground motion parameters (GMPs) or spectra such as response spectra (PSA) or Fourier spectra (FAS) for design events. With the increase in the number of strong-motion databases around the world (e.g., GeoNet, NGA-West2, and NGA-Sub), data-driven methods have become a popular approach to developing GMMs. The present thesis explores these methods to develop models to predict GMPs, PSA, and FAS with lower standard deviations. Further, these models are verified for their applicability to the Indian subcontinent. Initially, a hybrid network combining genetic algorithm and artificial neural network is applied to develop GMPs and PSA to a recently updated GeoNet database corresponding to the New Zealand region. The input parameters of the network are magnitude, Joyner-Boore distance, shear wave velocity, depth to the top of the rupture plane, and focal and tectonic flags. The developed networks are found to be robust and had lower standard deviations compared to existing models developed in this region. Northeast India has similar tectonics compared to the New Zealand region. Thus some of the significant earthquakes in this region are predicted using the developed network. It is concluded that the current models can be adapted in such regions to estimate ground motion. Later, FAS is predicted using the same inputs by coupling the hybrid network with an autoencoder to the GeoNet database. The developed network resulted in lower standard deviations than other popular FAS models. Additionally, an ensemble of time histories is generated from the predicted FAS. Further, predicting PSA from FAS using a long-short term memory (LSTM) network is proposed.