1

I have a large dataset that I need to predict some values for. I have this small reproducible example below. I would like to predict for a few years ahead using either gam or glm, however, I can't wrap my head around it since I don't have statistics background. Could some one guide me on how to predict and add the predicted values to my original dataset?

library(mgcv)
library(tidyverse)
m <- structure(list(year = c(2003, 2003, 2003, 2003, 2003, 2003, 2003, 
2003, 2003, 2004, 2004, 2004, 2004, 2004, 2004, 2004, 2004, 2004, 
2005, 2005, 2005, 2006, 2006, 2006, 2006, 2006, 2007, 2007, 2007, 
2007, 2007), month = c("August", "August", "September", "September", 
"October", "October", "November", "November", "December", "August", 
"August", "September", "September", "October", "October", "November", 
"November", "December", "August", "August", "September", "October", 
"October", "November", "November", "December", "August", "August", 
"September", "September", "October"), date = structure(c(12265, 
12281, 12296, 12311, 12326, 12342, 12357, 12372, 12387, 12631, 
12647, 12662, 12677, 12692, 12708, 12723, 12738, 12753, 12996, 
13012, 13027, 13422, 13438, 13453, 13468, 13483, 13726, 13742, 
13757, 13772, 13787), class = "Date"), turb = c(12.5, 11.8, 8.9, 
6.2, 5.1, 3.4, -0.1, 0, -0.1, 12.5, 10.4, 8.6, 6.1, 4.7, 1.4, 
0.1, -0.1, -0.1, 12.5, 11, 8, 4.4, 1.1, 0.5, -0.2, -0.2, 12.5, 
11.7, 10, 5.9, 3.6)), row.names = c(NA, -31L), class = c("tbl_df", 
"tbl", "data.frame"))

m$date <- as.numeric(m$date)
mod1 <- gam(turb ~ s(date), data = m)
summary(mod1)

#How can I predict 3 more years (to 2010) and have those predictions
#added to the bottom of the original dataset(m)?
Improve this question
2
  • 1
    Check out the predict function Commented Feb 9, 2024 at 16:51
  • 1
    You're hoping to predict date based on the value of turb, or the other way around? Commented Feb 9, 2024 at 16:52

1 Answer 1

Reset to default
1

Perhaps I'm wrong, but I think you're trying to predict turb based on the smoothed date. If so, then

m$datenum <- as.numeric(m$date) # just to keep `date` unchanged
mod1 <- mgcv::gam(turb ~ s(datenum), data = m)
new_m <- tibble(date = c(outer(max(m$date) %m+% months(0:36), c(0, 15), `+`))[-1]) %>%
  mutate(datenum = as.numeric(date))
new_m
# # A tibble: 73 × 2
#    date       datenum
#    <date>       <dbl>
#  1 2007-11-01   13818
#  2 2007-12-01   13848
#  3 2008-01-01   13879
#  4 2008-02-01   13910
#  5 2008-03-01   13939
#  6 2008-04-01   13970
#  7 2008-05-01   14000
#  8 2008-06-01   14031
#  9 2008-07-01   14061
# 10 2008-08-01   14092
# # ℹ 63 more rows
# # ℹ Use `print(n = ...)` to see more rows
predict(mod1, newdata = new_m)
#           1           2           3           4           5           6           7           8           9          10          11          12          13 
#   -1.246062   -6.392131  -11.709736  -17.027341  -22.001874  -27.319479  -32.465548  -37.783153  -42.929222  -48.246827  -53.564432  -58.710501  -64.028106 
#          14          15          16          17          18          19          20          21          22          23          24          25          26 
#  -69.174175  -74.491780  -79.809385  -84.612383  -89.929987  -95.076057 -100.393661 -105.539731 -110.857335 -116.174940 -121.321010 -126.638614 -131.784684 
#          27          28          29          30          31          32          33          34          35          36          37          38          39 
# -137.102288 -142.419893 -147.222891 -152.540496 -157.686565 -163.004170 -168.150239 -173.467844 -178.785449 -183.931518    1.498509   -3.819096   -8.965165 
#          40          41          42          43          44          45          46          47          48          49          50          51          52 
#  -14.282770  -19.600375  -24.574909  -29.892514  -35.038583  -40.356188  -45.502257  -50.819862  -56.137466  -61.283536  -66.601140  -71.747210  -77.064814 
#          53          54          55          56          57          58          59          60          61          62          63          64          65 
#  -82.382419  -87.185417  -92.503022  -97.649091 -102.966696 -108.112765 -113.430370 -118.747975 -123.894044 -129.211649 -134.357718 -139.675323 -144.992928 
#          66          67          68          69          70          71          72          73 
# -149.795926 -155.113531 -160.259600 -165.577205 -170.723274 -176.040879 -181.358483 -186.504553

This can be combined with the original data with

combined <- new_m %>%
  mutate(turb = predict(mod1, newdata = cur_data()), year = as.integer(format(date, format="%Y")), month = format(date, format = "%B")) %>%
  bind_rows(m) %>%
  arrange(date)
combined
# # A tibble: 104 × 5
#    date       datenum      turb  year month    
#    <date>       <dbl> <dbl[1d]> <dbl> <chr>    
#  1 2003-08-01   12265      12.5  2003 August   
#  2 2003-08-17   12281      11.8  2003 August   
#  3 2003-09-01   12296       8.9  2003 September
#  4 2003-09-16   12311       6.2  2003 September
#  5 2003-10-01   12326       5.1  2003 October  
#  6 2003-10-17   12342       3.4  2003 October  
#  7 2003-11-01   12357      -0.1  2003 November 
#  8 2003-11-16   12372       0    2003 November 
#  9 2003-12-01   12387      -0.1  2003 December 
# 10 2004-08-01   12631      12.5  2004 August   
# # ℹ 94 more rows
# # ℹ Use `print(n = ...)` to see more rows
tail(combined)
# # A tibble: 6 × 5
#   date       datenum      turb  year month    
#   <date>       <dbl> <dbl[1d]> <dbl> <chr>    
# 1 2010-08-01   14822     -173.  2010 August   
# 2 2010-08-16   14837     -176.  2010 August   
# 3 2010-09-01   14853     -179.  2010 September
# 4 2010-09-16   14868     -181.  2010 September
# 5 2010-10-01   14883     -184.  2010 October  
# 6 2010-10-16   14898     -187.  2010 October
Improve this answer
Sign up to request clarification or add additional context in comments.

9 Comments

I wonder why those predictions are exponentialy going negative to -187 @r2evans. My smallest turb value is -0.2 and my largest is 12.5.
My answer is demonstrating the use of predict(mod1, newdata=). I'm not advising on the usability of that GAM with your data. "Prediction is very difficult, especially if it’s about the future" (quoteinvestigator.com/2013/10/20/no-predict). If you want that, I suggest you migrate the question to Cross Validated where the theory of models is more appropriate.
You're trying to predict turb as a function of date, which, by the look of your original example, is not the case. You need to find the appropriate predictor(s) for turb in the original data set.
@Salvador, I think the answer addresses your question "predict more 3 more years and have those predictions added to the bottom of the original dataset". Regardless of the applicability of your actual model formulation, the mechanism is going to remain the same. Do you agree, or am I misunderstanding the question?
The default smooth will linearly extrapolate when forecasting outside the range of the data. This is unlikely to yield good predictions. You might like to try the MVGAM package or our MRFtools package (only on GitHub) to use a RW1 as the trend
|

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.