Multivariate Grey Prediction Models for Pattern Classification Irrespective of Time Series

Pattern classification can be regarded as a grey system problem because multiple factors of a pattern influence the class into which it can be categorized, but the relationship between these factors and the categorization is not clear. Multivariate grey prediction models (MGPMs), such as the GM(1, N), have thus drawn interest in pattern classification. However, as traditional MGPMs have been designed for time series forecasting, it is interesting to transfer each permutation in the collected data without involving temporal order to a time series. Any permutation among the patterns can also have a certain influence on classification performance. To solve this challenging problem, this study proposes several multivariate grey classification models by integrating genetic algorithms into multivariate grey prediction models. The results of experiments verified the usefulness of MGPMs for pattern classification.