Abstract: In the field of multiple attributes decision making, sorting decision has become an important kind of issue and been widely concerned in many practical application areas. In the process of making sorting decision,the rational and effective feature selection methods can extract informative and pertinent attributes, and thus improve the efficiency of decision making. From the extant literatures, many valuable researches have been provided for more reasonably solving this problem in the context of diverse data types, such as single value, null value and set value. However, very few studies focus on the sorting decision in term of interval-valued data. The objective of this paper is to provide a new feature selection approach for interval sorting decision by using the interval outranking relation. By integrating rough set model and information entropy theory, a new measurement called complementary condition entropy, which investigates the complementary nature of the relevant sets, is proposed for feature evaluation through analyzing the inherent implication of correlation between considered attributes in the problem of interval sorting decision. Furthermore,on the basis of the difference of the values of complementary condition entropy,the representation of the indispensable attributes and the measurement of attributes importance are presented, and then develop a heuristic feature selection algorithm is proposed for interval sorting decision. Finally, two illustrative applications, namely,the issues of venture investment and portfolio selection, are employed to demonstrate the validity of the proposed method.For the problem of multi-stage venture investment decision, through investigating the competitiveness, development capacity and financial capability of 16 investment projects, the corresponding probabilistic decision rules having better generalization capability, which can be used to determine whether to perform further investment. As to the issue of portfolio selection, 91 stocks coming from Chinese stock market and 9 operating performance indicators of these firms are employed. By using the presented approach in this study, a portfolio which has better investment return can be construeted. Accordingly, the corresponding strategy for building portfolio is useful to quantitative investment decision. In brief, as the important preprocessing tool in the process of decision analysis, the feature selection method built in this paper is of extensive meaning for discovering the key indicators and improving decision performance in the field of sorting decisio

Key words: interval-valued data, sorting decision, feature selection, complementary condition entropy