Skip to Main Content
Signals represented by their proper (in the sense of Nyquist-Shannon theorem) sampled values as time-series are commonly categorized by objective features such as amplitude and frequency distribution. The paper presents a novel signal classification method based on full difference expansion, mapping this expansion to qualitative space and extraction of subjective visual attributes from the qualitative space. The full qualitative difference expansion yields a vector that conveys total information on the variation of the particular signal represented by time-series and can be seen as a single point in n-dimensional discrete-space. From such a discrete-space, symbolic and numeric features are subjectively extracted and used for the decision tree construction that is consequently used in signal classification. The proposed method was tested in the context of the standard control chart pattern data, which are time-series used in statistical process control. The results are compared with other similar methods.