> ! \pJava Excel API v2.6 Ba==h\:#8X@"1Arial1Arial1Arial1Arial + ) , * `?DC,;4#titletitle[alternative]title[translated]author[google(nobreakline)] contributor[author(nobreakline)]"contributor[scopusid(nobreakline)]date[issued]relation[journaltitle]identifier[issn] citationsidentifier[major]subject(nobreakline) publisheridentifier[thesisdegree]contributor[advisor]relation[ispartofseries]relation[index(nobreakline)]typeabstractidentifier[govdoc]identifier[uri]identifier[doi(doi)]identifier[isbn]identifier[ismn]
identifier:New Sampling Approach to Zero-Inflated Count Data Analysis2020Y ĬYtTŐYP YMastert@ǽMaster's ThesisXCount data with excess zeros are common in various fields. However, the class imbalance problem of a data set has encountered a difficulty in predicting the response variables of a new data set. Conventional models Poisson and Negative Binomial model - tend to predict the probability of zero smaller than it is. Several modeling methods such as zero-inflated Poisson model or Poisson hurdle model have been proposed to address the imbalance problem of count data with excess zeros.
In this paper, the sampling-based method is proposed to handle zero-inflated count data. We will extend ROSE (Random Over-Sampling Examples) strategy to count data. ROSE was developed to mitigate the imbalanced binary data. With this extended ROSE strategy, we can generate zero deflated data and can make a better prediction. Simulation results show that the performance with this new strategy is better than the original modeling method in some zero-inflated data sets. It also has better performance for predicting fish counts with 56% zero proportion. It also takes less computing time than zero-inflated Poisson model.
;0 count pt0є , p, XY 䲑\ | } . 췘, pt0 8Ȕ \ pt0X Q | !XՔ p $D |0¨. |<\ count pt0X i xx xDš(Poisson) Ltm(Negative Binomial) Ӕ 0t , U`D T ǌ !XՔ t . 0|, 0 count pt0| \ x )`t H T, \<\ zero-inflated Poisson(ZIP) zero-inflated Negative Binomial(ZINB) .
|8, 0 count pt0 \ 0X )D HX. t pt0 H ROSE(Random Over-Sampling Examplements) 0D count pt0 XՌ U֥X. ROSE \ t pt0 t !%D X0 t H )<\, 0tX $Ƅ(oversampling), T(undersampling) 9@ SMOTE(Synthetic Minority Over-sampling Technique) 0X \Ĭx pt0 , pt0 8| Xଐ X.
t U֥ ROSE 0<\, 0 count pt0 t T @ !%D ɔ xD i ¤ଐ X. ¬tX \0 t 0D \ xt |ǀ 0 pt0 t, 0tX 0 pt0 H x ) T @ 1D 䲔 D . \, pt0 ȩD t m H )X 1D ଐ X. 56%X 0 D(D ɔ fish pt0 t, t m H ROSE 0t 0tX ) Q | !XՔ p Ǵ T \ !%D 䲔 D . \, RX system timeD DPX t )t ZIP x Ű t T Ȍ x䲔 D .khttp://dcollection.ewha.ac.kr/common/orgView/000000163749;
http://dspace.ewha.ac.kr/handle/2015.oak/253242;@Ce|(!CPrs BKm~ AJlm@!C,N?aXzgpDf
dMbP?_*+%" ,,??U
!
"
l>@
Root EntryWorkbook SummaryInformation(DocumentSummaryInformation8