To address the issues of fingerprint database vulnerability to environmental changes and low positioning accuracy in WiFi indoor fingerprinting localization, an indoor fingerprinting localization algorithm based on the weighted fusion of LightGBM, CatBoost, and XGBoost ensemble learning was proposed. In the offline phase, a Boxplot filter based on median absolute deviation was used to effectively filter out outliers from the fingerprint database. The missing values in the filtered fingerprint database were then imputed using KNN (K-nearest neighbors), ensuring the stability of the fingerprint database. In the online phase, three ensemble learning models: LightGBM (light gradient boosting machine), CatBoost (categorical boosting), and XGBoost (extreme gradient boosting): were combined. The chaotic adaptive JAYA algorithm was used to dynamically adjust model weights, constructing a weighted fusion coordinate prediction model. Experimental results show that the proposed algorithm achieves an average positioning error of 1.38 m, reducing the error by 6.52% to 37.7% compared to PSO (particle swarm optimization)-ELM (extreme learning machine), KNN, LightGBM, CatBoost, XGBoost, and KNN + XGBoost algorithms, providing a robust and accurate solution for indoor positioning.