In geoscience, particularly in fields like sedimentology and hydrogeology, the presence or absence of sand can be crucial for various analyses and interpretations. Although there are multiple researches that worked on distinguishing sand from non-sand, an essential gap is there to estimate the probability of sand and the associated uncertainty of the prediction. Sand probability measurements provide valuable information about the distribution, geometry, and connectivity of sand bodies within the reservoir. This information is critical for reservoir modeling, volumetric calculations, and production forecasting. The estimation of sand probability is achievable by introducing randomness using the non-deterministic machine learning models. The novel goal of the current study is to estimate the probability of sand facies as well as measuring the uncertainty in the prediction while perpetuating the high accuracy by introducing non-deterministic XGB model. The proposed approach is able to achieve an F1 Score of 89% in classifying sand. Additionally, the prediction uncertainty and sand probability (P10, P50 and P90) are also calculated in unseen or blind well by using the Seismic attributes.