Null Values Estimation Method Based On IOWA Operators And Application For Incomplete Information Systems

In modern society, with the database and network technology developingrapidly,incomplete or inaccurate data are also growing significantly. The incomplete data hasbrought information processing unreliable results, and even seriously affect the result ofprocessing. Therefore, aneffective,scientific and reasonable method to deal with theseincomplete data is becoming more and more important.The IOWA operator is a new aggregation operator which is put forward by a famousAmerican scholar R R. Yager and Filev in1999on the basis of the OWA operator. Thisoperator has obtain a extensive application in many areas, including decision-making,neuralnetwork, databasesystem, the fuzzy logic controller, the expert system, marketresearch,mathematical programming, etc. However, the IOWA operator have not yet applied to theincomplete information system ever before, a new kind of incomplete information system nullvalue estimation method is proposed in this paper, and applied to the c-band radio signalrecognition domain. In this paper, the main research work are as follows:1. The research of the current commonly used incomplete information system null valueestimation method:Mean Completer estimation, Combinatorial Completer estimation,K-Nearest Neighbor estimation, K-means Clustering estimation, Fuzzy K-means Clusteringestimation, Expectation-Maximization estimation, ROUSTIDA estimation. The advantagesand disadvantages of the seven kinds of estimation methods are analyzed.2. The paper put forward a kind of incomplete information system null value estimationmethod, set the aggregation results from IOWA operator as fill values of the empty values,also gives the concrete implementation algorithm, here referred to as “BIOWAE” algorithm.The estimation method not only considers the relevant information between objects, alsoconsiders the relevant information between attributes. This method has two main difficulties:1)The choice of relevant properties. At present it is difficult to find a unified formula to solvethe relevant properties in a different environment information system.2) The determination ofweight vector of the IOWA operator. The choice of weight vector will directly affect theaggregation result. To get a good estimate result, the suitable weight vector for theinformation system must be found. Then the new method and Mean Completer estimation,K-Nearest Neighbor estimation, K-means Clustering estimation, Fuzzy K-means Clusteringestimation, Expectation-Maximization estimation, ROUSTIDA estimation has carried on thecomparison and analysis with instance. At last we draw the conclusion that the method is agood method for estimating the null value. 3. Put the new estimation method applied in c-band radio signal identification. The paperput an improved signal recognition method based on original c-band signal identification thatwhich mainly processes the signals under identified and the known type ones in differentpartitions,each signal of known type is divided into three areas:1) The determined area: if thesignal under identification is in the certain determined area of a kind of signal, then this signalis considered to be this kind signal.2) the estimated area:if the signal under identification is inthe estimated area of a kind of signal, then we need to correct the feature vector of the signalunder identification by the BIOWAE algorithm and recognize again.3) the uncertain area: ifthe signal under identification in the uncertain area of a kind of signal, this shows that the typeof a signal under identification can’t be ensured or be a new one. Finally, the improvedrecognition method has proved the c-band radio signal recognition rate by actual data to acertain degree.