[R-br] Selecionar variáveis de uma matriz de correlação

[Rodrigo Muniz]

Caros,
Tenho uma dataframe com cerca de 1500 variáveis e gostaria de saber a
correlação entre essas variáveis, então gerei uma matriz de correlação.
Como muitas variáveis estão altamente correlacionadas, eu gostaria de
selecionar apenas as variáveis que tenham, por exemplo, no máximo uma
correlação de 0,3. Para isso eu utilizei a função findCorrelation do pacote
caret. Pela descrição, a função foi feita para isso, selecionar as
variáveis mais correlacionadas, considerando um limite de corte.
Como eu gostaria das variáveis menos correlacionadas, eu inverti minha
matriz de correlação, pelo seguinte comando:
as.matrix(sqrt((matrix.cor-1)^2), desta forma, a função selecionaria as
variáveis menos correlacionada.
No entanto, o resultado da função sempre me retorna bandas altamente
correlacionada. Por isso gostaria de fazer de uma forma alternativa. Segue
um pequeno exemplo para ajudar nas considerações.
Se alguém souber uma forma de programar para obter esse resultado, aceito
um direcionamento.
Desde já agradeço pela atenção.

require(caret)

filtro=c(0.7) #Este parâmetro foi criado para definir o liminar de corte,
como a #matriz está inversa, um valor de 0,7 que dizer que a correlação
máxima entre as #variáveis deve cer de 0,3.

matrix.cor <- as.data.frame(cor(bandas))# Gerar a matriz de correlação

banda.selecao=bandas[,findCorrelation(as.matrix(sqrt((matrix.cor-1)^2)),
          cutoff=as.numeric(filtro), verbose=FALSE)] # selecionar do
dataframe #original apenas as variáveis com fraca correlação.

bandas=structure(list(X1417 = c(0.161042, 0.186075, 0.202966, 0.211156,
0.172813, 0.167929, 0.175015, 0.200225, 0.178143, 0.182516, 0.162152,
0.169258, 0.14955, 0.143907, 0.161655, 0.164643, 0.157703, 0.160282,
0.177145, 0.169043), X1366 = c(0.352479, 0.376561, 0.372766,
0.39279, 0.36767, 0.368855, 0.361146, 0.390109, 0.373625, 0.383793,
0.371615, 0.375047, 0.363033, 0.341382, 0.372219, 0.375438, 0.359356,
0.372921, 0.375115, 0.36681), X469 = c(0.07749, 0.098493, 0.083867,
0.101454, 0.087098, 0.088587, 0.079879, 0.083187, 0.087261, 0.091197,
0.085591, 0.091496, 0.081759, 0.093462, 0.094817, 0.089399, 0.087421,
0.08966, 0.097024, 0.084687), X1549 = c(0.233143, 0.262691, 0.266292,
0.280087, 0.247662, 0.242888, 0.237161, 0.269429, 0.25326, 0.260258,
0.239942, 0.244771, 0.224855, 0.211901, 0.237555, 0.242471, 0.232943,
0.234706, 0.248407, 0.243808), X1424 = c(0.149535, 0.176297,
0.193145, 0.197999, 0.16312, 0.156738, 0.165102, 0.190595, 0.165941,
0.171179, 0.149601, 0.158143, 0.139422, 0.131351, 0.150353, 0.153668,
0.145654, 0.149609, 0.162426, 0.158941), X1957 = c(0.045334,
0.078347, 0.055058, 0.071466, 0.04449, 0.052894, 0.066396, 0.087972,
0.053841, 0.073682, 0.068851, 0.04738, 0.044667, 0.060964, 0.061176,
0.065722, 0.052965, 0.05974, 0.07101, 0.076109), X1848 = c(0.247498,
0.280584, 0.274104, 0.278935, 0.252584, 0.256848, 0.235115, 0.274966,
0.268281, 0.267282, 0.255826, 0.23335, 0.237109, 0.230515, 0.262255,
0.258836, 0.2387, 0.265159, 0.271541, 0.265838), X1537 = c(0.218761,
0.248475, 0.254953, 0.264756, 0.233748, 0.230395, 0.224644, 0.256149,
0.236666, 0.24462, 0.223847, 0.229429, 0.210074, 0.198872, 0.220081,
0.227677, 0.217309, 0.21968, 0.231893, 0.226282), X1621 = c(0.285036,
0.31607, 0.313035, 0.331176, 0.302131, 0.300669, 0.290153, 0.323815,
0.307194, 0.318355, 0.298653, 0.303098, 0.285467, 0.269884, 0.297272,
0.301428, 0.291555, 0.300164, 0.302502, 0.299489), X1602 = c(0.273662,
0.302662, 0.304277, 0.319991, 0.29179, 0.291713, 0.281382, 0.313998,
0.296813, 0.305874, 0.286391, 0.294462, 0.27532, 0.260575, 0.287217,
0.289802, 0.279721, 0.285657, 0.291856, 0.289307), X1865 = c(0.199668,
0.229541, 0.230852, 0.245059, 0.226876, 0.215756, 0.207998, 0.254432,
0.20988, 0.226653, 0.206045, 0.194866, 0.195796, 0.183774, 0.214694,
0.231647, 0.207081, 0.208682, 0.223025, 0.199142), X1476 = c(0.144097,
0.170145, 0.187837, 0.192702, 0.158221, 0.154393, 0.160075, 0.184072,
0.162413, 0.166724, 0.147404, 0.15375, 0.133576, 0.128008, 0.146151,
0.150793, 0.143929, 0.141969, 0.158288, 0.154279), X2034 = c(0.081108,
0.110882, 0.100852, 0.109914, 0.109893, 0.06942, 0.128299, 0.139094,
0.10447, 0.125726, 0.118003, 0.075258, 0.090784, 0.055216, 0.102326,
0.101315, 0.08443, 0.123796, 0.104067, 0.102786), X1460 = c(0.136055,
0.161915, 0.177241, 0.184844, 0.149423, 0.143712, 0.150499, 0.173385,
0.150807, 0.156626, 0.138878, 0.145282, 0.125618, 0.120778, 0.138968,
0.139961, 0.133014, 0.132243, 0.150618, 0.144916), X1236 = c(0.41637,
0.440245, 0.422308, 0.453948, 0.428499, 0.436163, 0.427196, 0.454866,
0.433779, 0.451874, 0.447265, 0.44747, 0.44684, 0.422101, 0.451307,
0.452207, 0.436043, 0.454355, 0.44641, 0.437131), X1913 = c(0.049649,
0.06384, 0.080675, 0.0761, 0.062756, 0.056635, 0.074517, 0.084222,
0.054835, 0.05488, 0.06389, 0.058421, 0.054827, 0.050249, 0.063544,
0.066275, 0.057604, 0.072578, 0.070545, 0.061034), X1885 = c(0.105368,
0.136086, 0.161411, 0.146339, 0.115904, 0.121671, 0.149059, 0.145293,
0.125417, 0.145147, 0.124081, 0.112018, 0.101411, 0.096761, 0.122112,
0.114244, 0.105535, 0.122283, 0.140519, 0.121402), X492 = c(0.080821,
0.102833, 0.088743, 0.10809, 0.093344, 0.096766, 0.083556, 0.089436,
0.091601, 0.100154, 0.089925, 0.096294, 0.087353, 0.097834, 0.102614,
0.09692, 0.096883, 0.094752, 0.102917, 0.089361), X1916 = c(0.037402,
0.063136, 0.08053, 0.071535, 0.059838, 0.0555, 0.060523, 0.079581,
0.044433, 0.055137, 0.055831, 0.054455, 0.054067, 0.055332, 0.062323,
0.0651, 0.05896, 0.075877, 0.069466, 0.064714), X1249 = c(0.416616,
0.440854, 0.423568, 0.455739, 0.428821, 0.436904, 0.42868, 0.455141,
0.432762, 0.45234, 0.447037, 0.448069, 0.445743, 0.420802, 0.455476,
0.453444, 0.43817, 0.454327, 0.444599, 0.437981)), .Names = c("X1417",
"X1366", "X469", "X1549", "X1424", "X1957", "X1848", "X1537",
"X1621", "X1602", "X1865", "X1476", "X2034", "X1460", "X1236",
"X1913", "X1885", "X492", "X1916", "X1249"), row.names = c(NA,
20L), class = "data.frame")
-- 
*Rodrigo A. Muniz*
Eng. Agrônomo. Ms Produção Vegetal (UENF)
Doutorando em Engenharia de Sistemas Agrícolas (ESALQ/USP)
E-mail - muniz.ra em usp.br
Cel (19) 98300-4333 (Pessoal)
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <http://listas.inf.ufpr.br/pipermail/r-br/attachments/20141009/d717d54a/attachment.html>