Muhammad Izzuddin Mahali, Jenq-Shiou Leu, Nur Achmad Sulistyo Putro, Setya Widyawan Prakosa, Cries Avian
This research introduces a novel method for indoor positioning by integrating Fuzzy C-Means clustering with deep learning autoencoders, leveraging Wi-Fi RSSI fingerprinting for precise location prediction and building and floor level classification within a unified framework. The approach combines the comprehensibility of Fuzzy C-Means (FCM) with the robust feature extraction capabilities of deep autoencoders through a dual-phase strategy: extensive offline training utilizing RSSI fingerprint data followed by real-time online position prediction. The model's performance was rigorously tested using 5-Fold cross-validation, examining its effectiveness both with and without the integration of FCM on various cluster numbers. Evaluation metrics, including Euclidean distance error for regression and accuracy, F1-score, precision, and specificity for classification effectiveness, were used for comparative analysis. The results were visualized through regression plots, confusion matrices, and learning curves, supplemented by t-SNE visualizations that provide profound insights into the feature space. These detailed analyses highlight the considerable promise of this integrated approach in enhancing the accuracy and reliability of indoor positioning systems. © 2024 IEEE.
National Taiwan University of Science and Technology, Dept. of Electronic and Computer Engineering, Taipei, Taiwan; Universitas Negeri Yogyakarta, Dept. of Electronics and Informatics Engineering Education, Yogyakarta, Indonesia; Universitas Gadjah Mada, Dept. of Computer Science and Electronics, Yogyakarta, Indonesia
