Development a case-based classifier for predicting highly cited papers

In this paper, we discussed the feasibility of early recognition of highly cited papers with citation prediction tools. Because there are some noises in papers’ citation behaviors, the soft fuzzy rough set (SFRS), which is well robust to noises, is introduced in constructing the case-based classifier (CBC) for highly cited papers. After careful design that included: (a) feature reduction by SFRS; (b) case selection by the combination use of SFRS and the concept of case coverage; (c) reasoning by two classification techniques of case coverage based prediction and case score based prediction, this study demonstrates that the highly cited papers could be predicted by objectively assessed factors. It shows that features included the research capabilities of the first author, the papers’ quality and the reputation of journal are the most relevant predictors for highly cited papers.

► This paper shows a quantitative support of early recognition of highly cited papers by accessible bibliometric indicators.
► The soft fuzzy rough set was used to eliminate the negative influences brought by noises in the actual citation process.
► The most relevant predictors for highly cited papers were extracted out.