In existing seismic design codes, prioritizing the horizontal component of the earthquake as the primary factor contributing to structural damage is common. However, recent studies on ground motion records show that vertical components also significantly impact structural damage. Moreover, in the recorded ground motion data for the Himalayan region, for some of the events (Example: 2015 Nepal earthquake aftershocks, 1999 Chamoli earthquake), the vertical component of the earthquake exceeds the horizontal component. Whereas IS1893:2016 suggests the vertical spectra to be obtained using a constant scale, which is two-third of horizontal spectra. While global codes like ASCE 7-16 and Eurocode-8 (1994-1:2004) have suggested period-dependent v/h (vertical-to-horizontal) ratios. The evidence from the recorded data and to comply with the global design practices, there is a necessity to revisit the constant v/h assumption in the Indian seismic design codes. Moreover, there are very few prediction models considering the vertical ground motion component in the Indian context. Considering this issue, the present work aims to study the characteristics of vertical ground motion by developing the nonparametric model for v/h spectral ratio model of 5% damped acceleration spectra for the Himalayan region. The recorded data available for the region and global data are combined using suitable flags to arrive at the regional-level prediction models. An extensive evaluation of residues was made as part of the work. We employed various machine learning techniques (ANN, RNN and LSTM) for this work and checked the efficiency of these techniques in capturing the complex trends shown by v/h spectral ratio with different source, path, and site characteristics. Out of the models considered, we found that LSTM-based model performed the best in capturing the known attenuation features. Results from this study can provide suitable guidelines to revise the existing codal practices for the study region.