An autoencoder-based indoor localization approach to reduce the class imbalance problem in fingerprint data

Abstract

The demand for indoor localization services for indoor environments, and indoor positioning based on fingerprinting has attracted significant interest due to its high accuracy. Working with data having class imbalance has led to biased localization results, with lower accuracy. In this paper, we presented a class imbalance problem solution using an autoencoder model with RSSI fingerprint data. The autoencoder architecture includes two phases. In the first stage, the autoencoder is trained using a large dataset of fingerprint samples, irrespective of their location labels. In the second phase, the data generated through the autoencoder are augmented with the original data. Further, we used two mathematical approaches such as KL- Divergence and Euclidean Distance, to evaluate the localization and checked the accuracy with classifiers, such as a KNN, SVM and RF. Experimental results are presented to ensure that the autoencoder-based indoor localization approach offers a promising solution to mitigate the class imbalance problem in fingerprint data by effectively reducing location error.