This study investigates the enhanced performance of FFT-based indoor positioning systems using various machine-learning models in dynamic environments. By transforming time-domain signals into the frequency domain, FFT filtering significantly improves the accuracy and reliability of indoor localization. The Decision Tree Regression (DTR) model, Random Forest (RF), XGBoost (XGB), Feedforward Neural Network (FNN), Support Vector Regression (SVR), and Linear Regression (LR) models were evaluated. Notably, XGB achieved the best results with an RMSE of 0.4234 m and an R2 value of 0.9899, followed by DTR with an RMSE of 0.4922 m and an R2 value of 0.9851. The FNN model also showed substantial improvement, achieving an RMSE of 1.3743 m and an R2 value of 0.8955. The study underscores the critical role of FFT filtering in enhancing indoor positioning accuracy and suggests that combining FFT with machine learning algorithms can lead to high-precision localization solutions. Future research could further explore the integration of FFT with other advanced models to advance the field of indoor localization.