Data clustering is a basic problem in pattern recognition, whose goal is grouping similar data into the same cluster. If original data are not well distributed due to large intra-variance, it would be difficult for traditional clustering algorithms to achieve satisfying performance. Thus, the auto-encoder network which is a good candidate to handle this problem is used for clustering of the received data. It provides a non-linear mapping function by iteratively learning the encoder and the decoder. The encoder is actually the non-linear mapping function, and the decoder demands accurate data reconstruction from the representation generated by the encoder. Simulation results show that the proposed clustering method can improve the neural network training surface to achieve the highest possible accuracy of the fingerprinting positioning method compared with conventional clustering method.

Keywords: auto-encoder, non-linear mapping function, clustering method