Abstract: Wi-Fi channel state information (CSI) contains rich feature information, which enables CSI based fingerprint positioning methods to build higher-dimensional features to improve positioning accuracy. However, the redundant information in the fingerprint features also makes the built-up fingerprint database large in storage, and increasing of the time cost of establishing the positioning model becomes large, and the real-time positioning calculation. In this regard, this paper proposes to use principal component analysis (PCA) method to reduce the dimensionality of the original fingerprint features, and then use the sequential minimal optimization (SMO) algorithm to establish the regression model of the reduced feature and the corresponding position and predict the position. The experimental results show that the algorithm in this paper can effectively overcome the above problems, while the average positioning error is 1.25 m, and the cumulative probability of positioning error within 2 m can reach 97%.