This work applies five machine learning algorithms—K-Nearest Neighbors (KNN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machines (SVM), and a Fully Connected Neural Network (FCNN)—to WiFi-based indoor localization using a subset of the SODIndoorLoc dataset, namely from the CETC331 building. We evaluate both scene-level and floor-level performance, highlighting the influence of environmental factors on localization accuracy, including obstacles, multipath propagation, and spatial layout. Experimental results indicate that KNN attains the minimal coordinate error (RMSE of 2.83 m), closely succeeded by FCNN and SVM, whilst RF and XGBoost demonstrate strong regression capabilities with \(R^2\) scores over 0.89. All models achieve nearly perfect floor classification, with a slight decline to 98.75% noted for XGBoost on a single floor. These findings provide guidance for algorithm selection according to the complexity of the indoor environment.