Structure and properties of grain boundaries (GB) for example tilt, twist, ∑3, ∑5, etc. significantly influence the microstructure and properties of the materials. Owing to the experimental difficulties, atomistic simulations are popular in studying these grain boundaries. However, the post processing and analysis of the simulation results is not straight forward and usually requiring expert’s intervention which may add to bias in analysis. Therefore, in this work we present a deep learning (DL) based framework for segmenting the grain boundary atoms from the bulk atoms in 3D dataset of atomic positions obtained from molecular dynamics (MD) simulations. Publicly available MD simulation dataset of Aluminum GB structures is used in this study. Raw MD simulation snapshots obtained from this dataset are pre-processed to extract relevant features to represent the various grain boundary configurations. Advanced machine learning (ML) algorithms like K-means and 3D convolutional neural network (CNN) are used on this dataset. K-means algorithm is used to extract classes in the dataset and later 3D CNN model was used for classification of bulk and grain boundary atoms for identification of grain boundary. The K-means clustering analysis revealed distinct patterns in the dataset for grain boundary and bulk atoms. The 3D CNN model significantly outperformed the conventionally used common neighbor analysis (CNA) method in identification of grain boundary atoms. As evident from case studies of different misorientation angles the 3D CNN achieved far lower misclassification rates, accurately distinguishing between grain boundary and bulk atoms. This work demonstrates the efficacy of advanced DL based model for accurate characterization of material features in atomistic simulations.