0

Hello everyone, I am using a neuralnet package in R. I am using three 'x' variables to predict a 'y' variable. I am dealing with the time series data with 82 observations. I divided the data set into training and test sets. Using 'compute' function I tested the neuralnet fit model on test data. However, I want to forecast into future periods with no x data at hand. I am trying to use 'forecast' function. Is there any way to use forecast function for neuralnet package in R. I appreciate any help in this regard. Find the R code below:

## Creating Training (1:72) and Test Slices (73:82)
library(caret)
timeSlices <- createTimeSlices(1:nrow(data1), 
                           initialWindow = 72, horizon = 10, fixedWindow = TRUE)

There are total 82 observations. I used createTimeSlices from caret package to create a training set (1:72) and test set (73:82). Now I use neuralnet packages to train the model as shown below:

library(neuralnet)

set.seed(101)

Model1 <- neuralnet(ppy ~ ppx1+ppx2+ppx3,data=mytraindata,
                hidden=4, threshold=0.001, stepmax = 2e+05, rep = 1, 
                lifesign = "full", lifesign.step = 1000, 
                algorithm = "rprop+", err.fct = "sse", 
                act.fct = "logistic", linear.output = TRUE)

Now I predict the model on test set (73:82), with the following code,

Predmodel1 <- compute(Model1, Ntestdata)

Until this point everything is cool. However, I need to forecast the future y variables. I tried to use 'forecast' function to predict y values for the future 7 periods (i.e., periods 83:89) as follows

library(forecast)
predmodel2 <- forecast(Model1$net.result[[1]], 7)

I got the following error:

Error in ets(object, lambda = lambda, allow. multiplicative. trend = allow. multiplicative. trend, : y should be a univariate time series

Any thoughts how to forecast y into 7 periods (83:89) into future? To forecast the future 7 periods, let us assume that the Model1 is trained on observations 1:82. I greatly appreciate your feedback. Thank you.

milos.ai
  • 3,882
  • 7
  • 31
  • 33
Raj
  • 11
  • 3
  • Welcome to Stack Overflow! Please update your code to include a reproducible example, namely runnable code that captures what you have done and what still needs to be done. You can read more about creating reproducible examples in R at http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example – josliber Nov 23 '15 at 05:42
  • As per your recommendation, I edited my questions along with the code. I appreciate your suggestions and feedback. – Raj Nov 23 '15 at 18:43

1 Answers1

0

forecast() Function is only for uni-variate data not for multivariate approaches

Panchacookie
  • 427
  • 1
  • 6
  • 22