Abstract: Aiming at the problems of heary work during establishment of traditional fingerprint database and the large matching error of the traditional random forest, an improved random forest localization method was proposed based on automatic fingerprint database expansion (FDE-IRF) to enhance the efficiency of fingerprint database construction and the accuracy of fingerprint matching. This method improved the traditional all sampling method to construct fingerprint database and the traditional random forest regression positioning method. The combination of sparse sampling fingerprint data of multiple time periods and Kriging interpolation method to complete unsampled fingerprint points improves the efficiency of database construction and gets a strong representative fingerprint database. At the same time, the decision tree weighting strategy is used to improve the average voting method in the traditional random forest, and the data out of bag is used to evaluate the prediction error of the decision tree and assign the corresponding weight, which improves the regression accuracy of the algorithm. The experimental results shows that the average positioning error of the proposed method is 1.26 m, which is at least 14.3% lower than that of similar methods, and verifies the accuracy and effectiveness of the proposed method.