Abstract: In terms of the traditional reserve prediction methods, such as Hubbert model and Weng's model, univariate polynomials are generally used to fit the reserve growth trend, which are unable to determine the influence of multiple variables on reserve prediction,resulting in a significant gap between the predicted results and objective reality. Based on the random forest machine learning model,a new method for predicting the growth trend of cumulative proven reserves is established, with the details as follows: Identify the quantifiable indicators that affect the growth of proven reserves through correlation analysis to determine the input attributes in the training model; Establish a random forest machine learning sample data set by taking the annual cumulative proven reserves of oilfields in the same basin as the evaluation unit; Optimize and train the model by adjusting the number of decision trees and the maximum characteristic number of a single decision tree, thus establishing a prediction model for the cumulative proven reserves, which supports to predict the nonlinear reserve growth affected by multiple factors. As a result, remarkable results have been achieved in predicting the annual cumulative proven reserves of oil fields in the eastern fault basin by applying this method, with a fitting accuracy of up to 88.19%, showing great promotion and application value.