Efficient Outlier Detection in RFID Trails
Elio Masciari, Giuseppe M. · 2009
In this chapter we addressed the problem of detecting outliers in RFID readings stream. The technique we have proposed is mainly based on the idea of representing an stream of readings as a time series. Thereby, the structural similarity between two series can be computed by exploiting the Discrete Fourier Transform (DFT) of the associated signals. Experimental results showed the effectiveness of our approach, with particular regard to some of the encoding schemes defined in the paper. The current work is subject to further significant extensions. As a matter of fact, the structural similarity between routes can be refined exploiting additional information on the RFID stream such as the actual distance among the reader. In our current implementation, the reader encoding function does not take into account semantic similarities between tags being scanned (i.e., objects belonging to the same category). However, precision could be improved by exploiting tagged object similarity techniques, such as, e.g., the exploitation of suitable ontologies. FFT-based distance measures, different from the one introduced in the paper could be used. Indeed, the FFT transformation contains lots of information about the contents of the original stream, and different distance measures could be more appropriate to exhibit such information. A further possibility to improve the proposed encoding schemes is that of defining a different strategy for dealing with the stream of readings in particular we plan to use more robust methods (e.g., non parametric-ones) for determining outliers, like those introduced in (Subramaniam et al., 2006). However, eventually new implementation has to be carefully studied in order to avoid inefficiency that are common in a system that has to deal with huge amounts of data.