Researches On Incomplete Information Processing Methods For Classification Decisions

Among practical applications, incomplete data is prevalent in many informationsystems. The reasons are traced from the diversification of the sources of data,interference, error, omission or human factors. The theories and techniques in rough setscan achieve relatively objective results on preferably maintaining those fundamentalcharacteristic and hidden rules. Thus it can provide better data support for differentclassification decisions.In this paper, based on methods in rough set theory, incomplete data processingalgorithms and techniques for given classification decision are researched. The twotypical applications are digitized file management and multi-source fusion recognition.The main contents are described as follows:On the one hand, oriented to the specific needs of human classified managements,the management for prison digitized records is selected as the instance. According to thebasic principles and description specifications of information systems, the prisonpersonnel records are digital archived corresponding to concerned entries. An improveddata filled method is designed based on ROUSTIDA in which the missing-fulfilloperation is completed by probabilistic similarity. Further, a knowledge reductionmethod depended on information entropy properties importance is proposed to refinedecisions which can simplify filled operation and avoid interference decisions broughtby the unnecessary missing items and redundant data. The above algorithms arevalidated by using simulated data sets.On the other hand, previous ideas are applied to multi-source fusion recognitiontask. Under the situations with wide differences of sources and complex interferencecontext, a unified processing framework is presented to deal with the incompletenessduring data preparation procedure. At learning phase, the information entropy is used toanalyze various conditions attributes’ importance and accomplish the relativeknowledge reduction. At online processing phase, combined with other mathematicaltools, the pretreatment is executed by multi-factor hierarchical data filled strategy toprovide the better data support for subsequent process.