Aiming at the problems of long time and low positioning accuracy of indoor positioning algorithm, an indoor positioning algorithm based on improved LightGBM (light gradient boosting machine) algorithm was proposed. The algorithm first preprocessed the data in the fingerprint database, and removed outliers and outliers through KNN (K-nearest neighborhood) algorithm to reduce environmental noise interference and improve data reliability. Next, the algorithm divided the sample set into training set and test set, and used LightGBM algorithm to model. At the same time, genetic algorithm was used to adjust the parameters in LightGBM algorithm, and the optimal parameters were found according to the fitness function to obtain the LightGBM + GA (genetic algorithm) coordinate prediction model. Finally, a prediction model was established according to the optimized parameters to realize coordinate prediction. The experimental results show that the algorithm improves the accuracy of WiFi positioning by 0.1 m compared with XGBoost (extreme gradient boosting, XGBoost) algorithm and 0.19 m compared with GBDT algorithm. In terms of positioning time, LightGBM + GA algorithm is 5.10 s faster than GBDT (gradient boosting decision tree) algorithm and 5.97 s faster than XGBoost algorithm, which has good practicability.