Abstract Peak ground motions and spectral accelerations estimated from the prediction equations are highly significant in earthquake hazard studies. Recently, these predictive relationships developed for higher-order parameters obtained paramount importance as they describe different ground motion characteristics. The northeastern region of India experiences extreme seismicity due to the Indian plate subduction under the South Asian plate. However, only a few ground motion prediction equations (GMPEs) are available for such tectonic environments due to insufficient ground motion data. In this regard, it is noticed that the tectonic environment experienced by New Zealand is similar to that of northeast India. So, in this paper, two GMPE models for New Zealand are developed with the help of the artificial neural network (ANN) technique using the GeoNet database. Model-1 corresponds to various higher-order parameters, whereas model-2 developed for spectral accelerations (Sa) between 0.01 and 5s. Further, these models are compared against global and region-specific GMPEs. The developed models shows good agreement with other GMPEs and the data but slightly over predicts at distances greater than 300 km. Additional consideration of site-to-site variability in the current models reduced the total standard deviations of model-1 by 19–22 % and model-2 by 20%–23 %. Further, the estimates of these developed models are compared with some of the significant earthquakes in northeast India, and from these results, it is concluded that the current models can be adapted in such regions to estimate ground motion.