Estimating how frequently a river is expected to flood is essential for developing watershed management plans, designing flood control infrastructure, and determining the flood risk of a property next to a river. The conventional method of achieving this involves looking at annual peak flows recorded by river gauges and fitting a probability distribution to determine the exceedance probability of river flows. To transfer the information from gauged sites to ungauged catchments, a Regional Flood Frequency Analysis (RFFA) is carried out which regionalizes the information using data from hydro-morphologically similar gauged catchments and provides flow estimates for various return periods. However, the performance of these RFFA models is greatly influenced by the choice of similar gauged catchments and the adequacy and consistency of their observed flows. In our study, we propose the implementation of Extreme Gradient Boosting (XGBoost), a machine-learning approach to estimate different return period flows for the ungauged catchments. The XGBoost is a decision-tree-based method and utilizes similar catchments attributes and observed flow quantiles as the conventional method, but it has a better capability of learning the relationship between the attributes. Overall, the proposed methodology produces promising results both in terms of accuracy and consistency in the estimated return period flows and can pave the way for the adoption of XGBoost as a viable alternate approach for regional flood frequency analysis.