Solute segregation at grain boundaries (GBs) has emerged as a promising strategy for alloy design. While most previous studies on GB segregation have assumed that solute atoms occupy substitutional sites within GBs, recent first-principles studies have revealed that undersized elements, such as Ni, Cu, and Fe, can also segregate to hollow interstitial sites in certain Al symmetric-tilt GBs. In this study, we employ hybrid molecular dynamics (MD)/Monte Carlo (MC) simulations to investigate the segregation behavior of Ni in Al coincidence-site lattice (CSL) GBs. The results demonstrate that Ni atoms preferentially segregate to interstitial sites within the kite-like GB structures in Al CSL GBs. Building on these findings, we develop a robust method to systematically identify potential interstitial sites at CSL GBs. This method consists of two main steps: detecting candidate interstitial sites and filtering them based on their structural properties. The identified interstitial sites show a distribution that is consistent with the results of hybrid MD/MC simulations. By applying this method to nanocrystalline alloys, we calculate the interstitial segregation energies, significantly improving the accuracy of GB segregation predictions. Furthermore, machine learning models using smooth overlap of atomic positions (SOAP) descriptors successfully predict these segregation energies. This study underscores the importance of GB interstitial segregation in enhancing our understanding of solute behavior and provides valuable guidance for the design of advanced metal alloys.