Neural Networks in RStudio - r

Question

I am trying to predict the variable price range which has 4 levels (0,1,2,3) with the neural network.
But the table with the results just have 2 lines, instead of 4.
I don't know why the table is not 4x4.
The idx only has ones and fours. And the pred only has 0 and 3.
Sometimes the table has 3 lines, but never 4. And most of the times has only 2.
Please help.
THAANK YOUU
#install.packages("rJava")
library(rJava)
#install.packages("xlsx")
library(xlsx)
#Read file of dataToModelset
setwd("C:/Users/Utilizador/Desktop/AE")
dataread <- read.xlsx("datasetDeleteOutliers.xlsx",1)
dataread <-datasetDeleteOutliers
dataread <-datasetCorrelation
dataread <-datasetMedian
dataread <-datasetNearestNeighbor
#
data <- data.frame(dataread)
str(data)
data$clock_speed=as.numeric(data$clock_speed)
data$m_dep=as.numeric(data$m_dep)
str(data)
# Search for binary variables
is_binary <- function(x) length(unique(x)) == 2
sapply(data, is_binary)
summary(data)#para ver o resumo das carateristicas dos dados
#data normalization
#z_score_function <- function(y) {(y - mean(y, na.rm=TRUE))/(sd(y,na.rm=TRUE))}
#normed <- apply(data,2, z_score_function)
#dataNormed <- data.frame(normed)
#normalization <- function(y) {
#return ((y - min(y)) / (max(y) - min(y)))
#}
# Min-Max Normalization
#data$ram <- (data$ram - min(data$ram)) / (max(data$ram) - min(data$ram))
#data$px_height <- (data$px_height - min(data$px_height)) / (max(data$px_height) - min(data$px_height))
#data$px_width <- (data$px_width - min(data$px_width)) / (max(data$px_width) - min(data$px_width))
#data$battery_power <- (data$battery_power - min(data$battery_power)) / (max(data$battery_power) - min(data$battery_power))
# Declare to R which variables are numeric and which are categorical/nominal to R
cols_to_factors <- c('blue','dual_sim','four_g','three_g','touch_screen','wifi')
dataToModel <- data.frame(lapply(data[,cols_to_factors],as.factor))
dataToModel <- cbind(data[,!colnames(data)%in%cols_to_factors],dataToModel)
rm(cols_to_factors)
## Reordering the columns
dataToModel <- dataToModel[,c(1,16,2,17,3,18,4:14,19,20,21,15)]
## Variable to coerce to integer
dataToModel$n_cores <- as.integer(dataToModel$n_cores)
dataToModel$price_range<-as.factor(dataToModel$price_range)
#IDs
dataToModel<-cbind(c(1:nrow(dataToModel)),dataToModel)
colnames(dataToModel)[1]<-"id"
str(dataToModel)
#MOdel Neural Networks
library(neuralnet)
library(dplyr)
# Make training and validation data
set.seed(123)
train <- sample(nrow(dataToModel), nrow(dataToModel)*0.7)
test <- seq(nrow(dataToModel))[-train]
train_N <- dataToModel[train,]
test_N <- dataToModel[test,]
# Binarize the categorical output
train_N <- cbind(train_N, train_N$price_range == '0')
train_N <- cbind(train_N, train_N$price_range == '1')
train_N <- cbind(train_N, train_N$price_range == '2')
train_N <- cbind(train_N, train_N$price_range == '3')
names(train_N)[23:26] <- c('level_0', 'level_1', 'level_2','level_3')
# Fit model
nn <- neuralnet(
level_0 + level_1 + level_2+ level_3 ~ ram+px_height+px_width+battery_power,
data=train_N, hidden = 4,lifesign = "full")
print(nn)
plot(nn)
comp <- compute(nn, test_N[,-22])
pred.weights <- comp$net.result
idx <- apply(pred.weights,1, which.max)
idx
pred <- c('0','1','2','3')[idx]
pred
table(pred, test_N$price_range)
accuracy_NN<-sum(pred==test_N$price_range)/nrow(test_N)
accuracy_NN