|
Abstract:
|
This paper introduces a set of signal processing features that identify customers suspicious of being involved in money service business (MSB) without a license. We extract features from customer's transaction history both in time-domain or transform-domain. Time-domain features, such as the average time separation between two consecutive check transactions, can be perceived and interpreted by human experts. However, transform domain features, such as Fourier and wavelet based features, proved to extract more important features for classification. In particular, our results proved that the features extracted using wavelet transform can better identify the MSB-like behaviors. Test of statistical significance on every extracted feature using t-test and ANOVA proved the statistical significance of the extracted measures.
|
