An Enhanced Deep Neural Network Approach for WiFi Fingerprinting-Based Multi-Floor Indoor Localization

WiFi fingerprinting based on the Received Signal Strength Indicator (RSSI) is a widely used technique for indoor localization. However, achieving the necessary precision for most smart applications has proven difficult due to various challenges, including the time-varying characteristics of indoor RSSI, device heterogeneity, and ambiguity in the positional fingerprints. To address these concerns, Artificial Intelligence (AI) techniques such as Machine Learning (ML) and Deep Learning (DL) have been used to classify and predict indoor locations. While these techniques have shown promising results, they face two significant challenges. First, their complex multilayer architecture requires extensive training iterations. Second, the distribution of the input layer fluctuates as the networks parameters are updated. In this study, we propose a WiFi indoor localization framework based on Recursive Feature Elimination with Cross-Validation (RFECV) and Deep Neural Network with Batch Normalization (DNNBN). The CV process is repeated multiple times to assess the RFEs generalization ability for optimal feature selection, and the BN algorithm is integrated into each layer of the DNN to ensure consistent activation value distribution and stabilization of the training process. Two datasets were used to assess the performance of the proposed RFECV-DNNBN method. The results show that the method addresses the challenges faced by existing AI techniques and outperforms current methods in classification and regression tasks. Our proposed framework achieved 100% accuracy in building classification and 94.69% and 96.39% in floor classification on the UJIIndoorLoc and UTSIndoorLoc datasets, respectively, demonstrating its effectiveness. Additionally, it achieved a Mean Absolute Error (MAE) of 5.08 m and 4.29 m for the respective datasets, highlighting its potential in multi-floored environments.