本文關(guān)鍵詞：基于特征提取的多指標(biāo)面板數(shù)據(jù)聚類方法　出處：《統(tǒng)計(jì)與決策》2016年19期 　論文類型：期刊論文

  更多相關(guān)文章： 面板數(shù)據(jù) 特征提取 動(dòng)態(tài)聚類 交通事故

【摘要】：多指標(biāo)面板數(shù)據(jù)能夠較全面的提供研究對(duì)象的信息和數(shù)據(jù)特征,但復(fù)雜的數(shù)據(jù)結(jié)構(gòu)也給其聚類分析帶來了一定的困難。針對(duì)這一問題,文章提出了基于特征提取的多指標(biāo)面板數(shù)據(jù)聚類方法,該方法將能夠表征面板數(shù)據(jù)動(dòng)態(tài)變化的"絕對(duì)量"特征、"波動(dòng)"特征、"偏度"特征、"峰度"特征及"趨勢(shì)"特征引入動(dòng)態(tài)聚類算法中,可以避免以往采用歐式距離進(jìn)行聚類的局限性,還可以處理帶有缺失數(shù)據(jù)的面板數(shù)據(jù),同時(shí)大大提高了聚類效率,并最大限度地保證時(shí)間維度信息不受損失。利用該方法分析了2001至2013年我國不同省份道路交通事故的不平衡狀況,通過實(shí)證分析表明該方法能夠解決多指標(biāo)面板數(shù)據(jù)聚類的問題。
[Abstract]:Multi-index panel data can provide more comprehensive information and data features of the research object, but the complex data structure also brings some difficulties to its clustering analysis. In this paper, a multi-index panel data clustering method based on feature extraction is proposed, which can represent the features of "absolute quantity", "fluctuation" and "bias" of the dynamic change of panel data. "kurtosis" feature and "trend" feature are introduced into dynamic clustering algorithm, which can avoid the limitation of using Euclidean distance to cluster in the past, and can also deal with panel data with missing data. At the same time, the clustering efficiency is greatly improved, and the time dimension information is guaranteed to be free from loss to the maximum extent. Using this method, the unbalanced situation of road traffic accidents in different provinces of China from 2001 to 2013 is analyzed. The empirical analysis shows that this method can solve the problem of multi-index panel data clustering.
【作者單位】： 南京航空航天大學(xué)經(jīng)濟(jì)與管理學(xué)院;
【基金】：國家自然科學(xué)基金資助項(xiàng)目(71071077,71371098) 中央高�；究蒲袠I(yè)務(wù)費(fèi)專項(xiàng)資金資助項(xiàng)目(NC2012001) 南京航空航天大學(xué)基本科研業(yè)務(wù)費(fèi)專項(xiàng)科研資助項(xiàng)目(NZ2010006)
【分類號(hào)】：F224
【正文快照】： 0 引言 面板數(shù)據(jù)又稱時(shí)間序列截面數(shù)據(jù)或混合數(shù)據(jù),是由不同個(gè)體在多個(gè)時(shí)期上觀測得到的數(shù)據(jù),它綜合了時(shí)間序列與截面數(shù)據(jù)的特征,能夠描述研究對(duì)象的動(dòng)態(tài)特征,在經(jīng)濟(jì)與管理問題的建模與分析中具有重要的地位。面板數(shù)據(jù)的聚類分析是指依據(jù)不同指標(biāo)的觀測數(shù)據(jù),把數(shù)據(jù)劃分為若干

【相似文獻(xiàn)】

相關(guān)期刊論文 前1條

1 蔣輝;張波;;支持向量回歸特征提取的ARMA準(zhǔn)則——中國社會(huì)消費(fèi)品零售總額預(yù)測的實(shí)證研究[J];統(tǒng)計(jì)與信息論壇;2012年07期

，

本文編號(hào)：1382130