Research And Application Of Missing Data Processing Method

In the current research and investigation in each field,data missing scenarios often happens due to various kinds of known or unknown factors. It will not only increase the difficulty of the researchers'analysis of data, but will also cause the deviation of the results, thus reduce the efficiency of he researchers'statistical work. So how to eliminate or minimize the impact of these missing data becomes more and more important. To effectively solve the problems caused by missing data, the author fills the missing data to form a complete data set firstly, and then analyzes statistically the complete data sets correspondingly.This paper researches the methods of handling missing data, from deleting individual cases, weighted, imputation and model based methods. The researchers introduce KNN, EM algorithm and BP Neural network imputation in detail and analyze their difference. Meanwhile, then the researchers apply the SVM and control-points optimization to handle the missing data. Finally, functional model has been made based on the characteristics of SVM in function fitting. After the main factors and data set are selected, the model is trained by the data set, and then successfully used to complete the missing values. It has proved that the method of completing missing values based on SVM and control-points optimization smoothing method is practical.