Determination Methods Of Disaster Assessment Variables And Weights Under Missing Data

Quantification, standardization and modeling are research direction of disaster assessment; existing mature assessment methods usually are established depended substantial data. And disaster assessment follows a process which includes analyzing disaster data, selecting assessment variables, determining variable weights, establishing assessment model and obtaining assessment results. The first three procedures are classified data preprocessing topic, and they influence the effectiveness of assessment models and the accuracy of assessment results directly.It is necessary to consider the influence of missing disaster data, and in the basis of existing disaster assessment, how to determine assessment variables and variable weights in incomplete disaster data situation is the critical research problem to assist the establishment of assessment model. On one hand, missing data is the main characteristic of disaster data, but complete data is the basic requirement to construct assessment model. Thus how to deal with missing data is the first problem to be solved. On the other hand, people know a little about the disaster mechanism because of the complexity of nature disaster, and most of the assessment models are built based on statistical analysis and law mining of the disaster data instead of mechanism inference. And according to analyze and resolve disaster assessment models, assessment variables and variable weights are the common and only parameters in all assessment models. The scientific and effectiveness of assessment models depend on defining and evaluating the two types of parameters, thus select assessment variables and determine variable weights are critical factors in the process of building assessment models.Therefore, for the above problems, this paper did research based-on existing research achievement, and has made the following progress:(1) Analyze characteristic of missing disaster data and establish description model of disaster assessment. Analyze the reasons, classification and behavior of missing data, and then establish the description model of disaster assessment and representation model of assessment variables based on common knowledge of model.(2) Study the imputation method of missing disaster data. In the basis on analyzing missing behavior of disaster data, presented two imputation methods for discrete missing situation and continuous missing situation respectively. For discrete missing situation, proposed an imputation method based on k nearest neighbor algorithm by evaluating the relevance of disaster. For continuous missing situation, proposed the other imputation method based on scenario match algorithm; and according to defining "scenario", this method used the data of historical scenario which has the maximum suitability with the current scenario to supplement current data.(3) Study selection methods of assessment variables. Under the premise of complete data, we classified the selection methods to two types by means of analyzing the relationship between assessment variables and assessment objectives, namely selection method under single objective and selection method under multiple objectives. A method based on grey relation theory is proposed for single objective situation, and it only consider the relationship between disaster data and assessment objective. A two-stage process is established to select assessment variable for multiple objective situation taking into account subjective and non-subjective objectives.(4) Study weighting methods of assessment variables. According to the existing form of assessment variables in models, we presented two weighting methods for and sets of assessment variables. For single assessment variable, a weighting method based on information entropy was given by measuring the average information amout of assessment variables. For assessment index, according to quantify the relationship between variable weights and index weights, the research problem is converted to calculate correlation coefficient between variable and index. And then a weighting method based on correlation is established determine index weights.This paper focuses on selection and weighting methods of disaster assessment variables in the background of missing disaster data, and it is a useful exploration to support the establishment research of assessment model by completing disaster data and mining valuable assessment information of disaster data. The achievements of this paper partly enrich and improve the research blank of disaster data preprocessing field, and it also provides some research foundation for further researches.