Drilling and blasting are among the most efficient techniques for the development of headings in underground mines or tunnelling projects. The common drilling pattern used for smaller dimension headings in various cases is burn cut pattern. This pattern is associated with one major problem: insufficient pull. It is important to maximise pull in drivage blasts to reduce project costs and enhance efficiency. The maximisation of pull can be effectively achieved by predicting it prior to the execution of blast. Therefore, this study has been carried out to predict the percentage of pull obtained during drivage blasts in an underground metalliferous mine using machine learning algorithm. Specifically, random forest and k nearest neighbour algorithms were used for this purpose. Five parameters namely number of holes, average hole depth, explosive per hole, maximum explosive weight per delay and total explosive fired were collected from the experimental site for model development. A total of 146 datasets were collected and split into 70:30 ratio for training and testing purposes. The performance of predictive models was assessed using RMSE and R2 values. The RMSE and R2 obtained for the KNN model were 3.411 m and 0.74 respectively, while for the RF model, they were 1.731 m and 0.907 respectively. This indicates that the accuracy of pull prediction using RF model is higher as compared to the KNN model. Therefore, an RF based predictive model can be effectively used for the prediction of pull at the experimental site.