Aiming at the problems such as the low algorithm robustness and easily blurred fingerprint of channel state information (CSI) in indoor localization, a location algorithm based on locally linear embedding (LLE) and gradient boosting decision tree (GBDT) is proposed. In the offline stage, first, the preprocessed amplitude and phase are regarded as joint CSI fingerprints, and then the individual subcarriers of the joint fingerprint are weighted using the elastic network (EN) algorithm before dimensionality reduction by LLE, which not only ensures the authenticity of CSI fingerprint after dimensionality reduction, but also enhances its unique characteristics. Finally, GBDT algorithm based on fruit fly optimization algorithm (FOA) is used to train the reduced dimension data to improve the reliability and stability of CSI fingerprint, and the fingerprint database is established. In the online stage, the LLE+GBDT algorithm is adopted to find the fingerprint information of the test point, so that actual physical locations can be predicted by matching with the fingerprint library. The indoor localization experiments results show that the proposed algorithm has higher localization accuracy and robustness compared with the comparison algorithm, and has certain application value.