R does not recognize column names as predictors-CodePudding

I`m trying to build a random Forest model for predicting Arsenic concentration using sentinel 2 bands and vnir and mnir spectra as predictors. The model is constructed as follows:

RF_RAW <- randomForest(Arsenic ~  ., data = merged[, c(5, 10:1944)], importance= TRUE, na.action=na.omit )

The problem occurs with names of columns, as the function does not recognize it

Error in eval(predvars, data, env) : object '400' not found

The same happens for any other wavelength if I choose to omit 400, both vnir and mnir spectra, columns of Sentinel bands: B1 to B12 work, and RF works with just them. I can't figure out

    head(merged)
# A tibble: 6 x 1,944
  Sample_ID   Corg   H2O   KCL Arsenic Phospate   FID Longitude Latitude     B1     B2     B3     B4     B5    B6    B7    B8
  <chr>      <dbl> <dbl> <dbl>   <dbl>    <dbl> <dbl>     <dbl>    <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl> <dbl> <dbl> <dbl>
1 P1_S1_1    1.16   5.6   4.32  47958.     597.    10      16.9     50.4 0.0371 0.0358 0.0527 0.0557 0.101  0.204 0.224 0.219
2 P1_S1_2    0.470  6.24  4.7   50720.     398.     9      16.9     50.4 0.0371 0.0395 0.0606 0.0641 0.101  0.204 0.224 0.259
3 P2_S1_1    1.33   6.28  4.91  31055.     316.     8      16.9     50.4 0.0323 0.0436 0.0682 0.0729 0.107  0.182 0.226 0.232
4 P2_S1_2   12.7    7.67  7.07  37492.     695.     7      16.9     50.4 0.0322 0.034  0.0576 0.0422 0.102  0.223 0.231 0.288
5 P3_S3_1    0.617  6.76  5.64  54245.     249.     6      16.9     50.4 0.0322 0.0438 0.0662 0.0696 0.113  0.213 0.253 0.233
6 P4_S4_1   20.8    4.41  3.21   7175.     731.    11      16.9     50.4 0.027  0.0253 0.0554 0.0272 0.0966 0.281 0.356 0.404
# ... with 1,927 more variables: B8A <dbl>, B9 <dbl>, B11 <dbl>, B12 <dbl>, 400 <dbl>, 401 <dbl>, 402 <dbl>, 403 <dbl>,
#   404 <dbl>, 405 <dbl>, 406 <dbl>, 407 <dbl>, 408 <dbl>, 409 <dbl>, 410 <dbl>, 411 <dbl>, 412 <dbl>, 413 <dbl>, 414 <dbl>,
#   415 <dbl>, 416 <dbl>, 417 <dbl>, 418 <dbl>, 419 <dbl>, 420 <dbl>, 421 <dbl>, 422 <dbl>, 423 <dbl>, 424 <dbl>, 425 <dbl>,
#   426 <dbl>, 427 <dbl>, 428 <dbl>, 429 <dbl>, 430 <dbl>, 431 <dbl>, 432 <dbl>, 433 <dbl>, 434 <dbl>, 435 <dbl>, 436 <dbl>,
#   437 <dbl>, 438 <dbl>, 439 <dbl>, 440 <dbl>, 441 <dbl>, 442 <dbl>, 443 <dbl>, 444 <dbl>, 445 <dbl>, 446 <dbl>, 447 <dbl>,
#   448 <dbl>, 449 <dbl>, 450 <dbl>, 451 <dbl>, 452 <dbl>, 453 <dbl>, 454 <dbl>, 455 <dbl>, 456 <dbl>, 457 <dbl>, 458 <dbl>,
#   459 <dbl>, 460 <dbl>, 461 <dbl>, 462 <dbl>, 463 <dbl>, 464 <dbl>, 465 <dbl>, 466 <dbl>, 467 <dbl>, 468 <dbl>, ...,

colnames(merged)
   [1] "Sample_ID" "Corg"      "H2O"       "KCL"       "Arsenic"   "Phospate"  "FID"       "Longitude" "Latitude" 
  [10] "B1"        "B2"        "B3"        "B4"        "B5"        "B6"        "B7"        "B8"        "B8A"      
  [19] "B9"        "B11"       "B12"       "400"       "401"       "402"       "403"       "404"       "405"      
  [28] "406"       "407"       "408"       "409"       "410"       "411"       "412"       "413"       "414"      
  [37] "415"       "416"       "417"       "418"       "419"       "420"       "421"       "422"       "423"      
  [46] "424"       "425"       "426"       "427"       "428"       "429"       "430"       "431"       "432"

and so forth. https://drive.google.com/file/d/1xifstUBv6sqa8-c51ukRsw9oZyKtzK03/view?usp=sharing is csv merged used in this example. I have similar ones with applied spectral transformations and they behave the same.

Thanks in advance

CodePudding user response：

Looks like randomForest doesn't like numeric column names, fix the column names. Try this example:

# example data
x <- mtcars[1:10, 1:3]
x[, "400"] <- x$disp

library(randomForest)
colnames(x)
# [1] "mpg"  "cyl"  "disp" "400" 

# as expected we get error:
randomForest(mpg ~ ., data = x)
# Error in eval(predvars, data, env) : object '400' not found

# now fix the column names
colnames(x) <- make.names(colnames(x))
colnames(x)
# [1] "mpg"  "cyl"  "disp" "X400"

randomForest(mpg ~ ., data = x)
# Call:
#   randomForest(formula = mpg ~ ., data = x) 
# Type of random forest: regression
# Number of trees: 500
# No. of variables tried at each split: 1
# 
# Mean of squared residuals: 5.218906
# % Var explained: 31.39