Casting of steel alloys with CO₂ sand moulds is the most common yet very important manufacturing process. Various kinds of valve bodies are cast for heavy and light duty fluid applications. Many casting defects like shrinkage porosity, cracks, etc. arises during solidification that ultimately lead to rejections. Cooling rate (CR) and so the interfacial heat transfer coefficient (iHTC) becomes most important parameters in order to control and monitor such defects. In the present study, we compare the results of Machine Learning (ML) algorithms to predict closely to experimental data of the CO₂ mould sand casting of ASTM A487 4C. Various machine learning algorithms like Bayesian Ridge, Support Vector Regressor, Random Forest and K-Nearest Neighbor (KNN) is used with different sets of learning and testing data. Experimentally, iHTC was found rapidly decreasing initially but slowdowns rate during solidification. Random forest and KNN regression model predicted better results than other two but KNN found superior in prediction. Further, KNN with 2-neighbors was seen giving better predictions with minimum error rate than with any other number of neighbors. It is seen that if enough and accurate experimental data is collected, ML models give reliable results replacing a need of costly casting simulation softwares at certain level in defect prediction or optimizing parameter.