Gamma Ray Astronomy at PeV EnergieS-phase 3 (GRAPES-3) is a cosmic ray experiment with an array of extensive air shower detectors and a muon telescope. The primary goal of the experiment is the precision study of the cosmic ray energy spectrum, its nuclear composition and also multi-TeV $\gamma$-ray astronomy. The punch-through hadrons produced near the air shower core can lead to problems in the precise estimation of the number of muons and hadrons which is an essential parameter for reconstruction. Machine learning (ML) can prove to be immensely useful in distinguishing between different particle types which will significantly improve the physics analysis of the GRAPES-3 experiment. In this work, we have tested the feasibility of using Boosted Decision Trees (BDTs) for the task of muon-hadron separation at GRAPES-3. We study the efficiency of BDTs for separating muons from hadrons in extensive air showers detected in the experiment. We have obtained 89.5 % accuracy in classifying single incoming muon and hadron.