This study is being performed as there has been lack of studies in analyzing and predicting the seismic responses of asymmetric RCC buildings equipped with base isolator using machine learning (ML) approaches. This paper accounts the prediction of the top floor displacement of asymmetric fixed based and lead-rubber bearing, also known as New Zealand (NZ) bearing, base-isolated RCC buildings modeled as torsionally coupled structure due to the superstructure eccentricity using artificial neural network (ANN) and convolutional neural network (CNN) approaches. An efficient metamodeling technique is used to accurately account the displacement of the top floor of the RCC building under a horizontal earthquake-induced ground motion. A SDOF system is considered with unidirectional horizontal eccentricity. The parameters are limited to uncoupled frequencies in x direction, frequency ratios and eccentricity ratios of the superstructure and isolator used, along with isolator parameters such as isolation damping ratio, isolation time period, and normalized yield strength. This method is implemented in the time domain to accurately incorporate the time step dependent response prediction. For this, 2000 numbers of known structural response data are generated using the Latin Hypercube Sampling method. Training, validation, and testing are done with the metamodel taking 80% of generated data for training and validation and 20% of generated data for testing using ANN and CNN approaches. Then responses of the building are predicted for different time periods. The numerical results show that the application of CNN produces more accurate response prediction than the ANN.