How to implement a neural network in R language?

Question

Neural networks

It is necessary to implement a neural network in the R language to predict the size of the pension depending on the average salary. For implementation, I use the neuralnet package. As an environment - the program Rgui for windows.
First, we have two data series - the average salary in the city for the last 10 years and the average pension for the last 10 years. This data is used to train the neural network:

#средняя зарплата за каждый год
traininginput <- c(0.225, 690, 2313, 2931, 4061, 4937, 5809, 7096, 8803, 10095, 12229, 13572)
#средняя пенсия за каждый год
trainingoutput <- c(0.118, 274, 949, 1270, 1668, 2001, 2434, 3028, 3393, 4519, 5594, 7610)

I set data for training:

trainingdata <- cbind(traininginput,trainingoutput)
colnames(trainingdata) <- c("Input","Output")

And teaching her

net.pension <- neuralnet(Output~Input,trainingdata, hidden=10, threshold=0.01)
print(net.pension)

And now I want that when I enter the average salary for the next year, the neural network gives me a forecast of the average pension for the next year. How can this be done in R?
Now I do like this (but nothing works this way).

#Отправляем на вход среднюю зарплату на будущий год
testdata <- c(15851) 
net.results <- compute(net.pension, testdata) 
 
ls(net.results)
 
#Lets see the results
print(net.results)

Thanks for the help.

Answer 1

First of all, your data must be normalized for use in neural network training. Besides,

hidden=10

seems redundant to me.
Try this code:

library(neuralnet)

# 1. creating the initial data, plotting
data <- data.frame (
  input = c(0.225, 690, 2313, 2931, 4061, 4937, 5809, 7096, 8803, 10095, 12229, 13572),
  output = c(0.118, 274, 949, 1270, 1668, 2001, 2434, 3028, 3393, 4519, 5594, 7610)
)
plot(data$output ~ data$input, main="Distribution of the pension relative to the salary", xlab="Salary", ylab="Pension")

# 2. normalizing the data, plotting
min.input <- min(data$input)
min.output <- min(data$output)
range.input <- diff(range(data$input))
range.output <- diff(range(data$output))
data.norm <- data.frame (
  input = (data$input - min.input) / range.input,
  output = (data$output - min.output) / range.output
)
plot(data.norm$output ~ data.norm$input, main="Distribution of the pension relative to the salary (normalized)", xlab="Salary", ylab="Pension")

# 3. neural network
net <- neuralnet(output ~ input, data.norm)

# 4. test the output
testdata <- seq(0, 25000, by=500)
testdata.norm <- (testdata - min.input) / range.input
result <- round(compute(net, testdata.norm)$net.result * range.output + min.output)
plot(testdata, result, main="Predicred outcome", xlab="Salary", ylab="Pension")