On the appropriate pattern frequentness measure and pattern generation mode: a critical review

The classic case pattern mining is a fundamental subject in data mining and big data science. The goal of the mining is to find correctly from a given dataset the patterns and their respective intrinsic frequentness. This paper examines two important yet misused instruments, the pattern frequentness measure "support" and the full enumeration pattern generation mode, which cause serious Overfitting thus deviate from the mining goal. A theoretic combined solution for the two critical issues is then proposed. This solution plus the equilibrium condition introduced in this paper forms a set of three fundamental rationality check criteria that every mining approach should observe. As such, the rationality of the mining theory and the reliability of the mining results would be substantially improved from the previous work. These together promise a significant change towards more effective pattern mining.