Steam flood injection is a method based on thermal recovery that injects steam, or especially water, at particular temperatures through the special injection wells. In this research, the oil field injected in steam flood is a heavy oil maturation with high viscosity, while the preliminary prioritization in steam flood injection in an oil field must be done at certain stages in a particular area in this oil field. This area consists of 8 sub-areas, 1,2,3 north section of the sub-area and 1,2,3,4,5,6,7,8 south section of the sub-area, each of which has a different number of wells and subsurface conditions. The provision to prioritize the injection in certain areas consists of monthly well production data, well location data, and reservoir data. On reservoir data is used a machine learning method that is random forest regression, which aims to generate variable importance, which will be made by grading reservoir properties variables on which are the most important or relevant for use as a parameter to determine the preliminary prioritization. The use of this Random Forest Regression method was combined to get an accuracy score for high-level modelling and correlate the process and results with the original state of on-field development and reservoir properties. The flow of this method is data cleansing, fitting models to data, assessing the quality of fit, generating decision trees, and identifying key variables. The tools used for this method include a programming language, which will do processing and build the variable importance model in the Random Forest Regression method. This can be done in a systematic and structured way. The result of this method shows the variable importance of reservoir properties such as porosity, saturation, and permeability in the form of OOB value calculation. While the final result of all corresponding prioritizations indicates that the 4,5,6 south section sub-areas should be prioritized for steam flood injection first.