Research on the Preprocessing Algorithms of Flight Data Mining

With regard to the characteristics of flight data,outlier elimination and feature selection are studied respectively as the two important sides of flight data preprocessing.Two new methods are proposed,one is the outlier elimination method based on error detection,and the other is the two-phase feature selection method using neural network.The error vector between the initial sequence and the smooth sequence is used to judge outliers in the outlier elimination method,when the error is over the detection threshold.In the two-phase feature selection method,the feature importance is ranked by neural network,and then the important features are selected using neural network.Finally the effectiveness of the two methods is shown by the test data.