Abstract: For the time-varying fluctuation of wireless local area networks received signal strength in indoor complex environment, an indoor localization method based on the improved random forest algorithm and fused particle filter is proposed.Firstly, a decision tree is built by selecting the signal access points with smoother received signal strength as the localization fingerprint features through adaptive sliding windows, so as to reduce the noise in the feature space.Then, the decision tree is given different voting weight levels by using the out-of-bag data after bagging sampling to increase the discriminative variability among the decision tree categories, so as to optimize the localization accuracy of the improved random forest algorithm in the scenario of weak mapping correlation between target location and received signal strength.The problem of low localization accuracy in the scenario of weak mapping correlation between target location and received signal strength is optimized, and the mapping model between location coordinates and received signal strength is established based on the improved random forest algorithm to solve the location of the localization point.Finally, in order to improve the accuracy and stability of localization, the positions solved by the improved random forest algorithm is used as the observed quantity of particle filtering, and the target positions are estimated and predicted by particle filtering.The experimental results on the Zenodo dataset show that: the proposed algorithm achieves 75.8% accuracy in the range of 2 m positioning error, and the accuracy is already as high as 94.8% in the range of 4 m, compared with the traditional random forest algorithm, it can well improve the positioning accuracy in the range of 2 m by 11.5% and compared with indoor positioning algorithms such as K-nearest neighbors, ridge regression, and Bayesian, the positioning accuracy and algorithm real-time performance are also significantly.