Null Result in R data mining in Decision Tree

Question

I have this code

#Import data
tugas=read.csv("D:/FlightDelays.csv")
dipakai=c(1,2,4,8,10,13) 
l=dim(tugas)[1] 
tugas<-tugas[1:l,dipakai] 

## Pembagian Data TRaining dan testtin
n <- round(nrow(tugas)*0.70);n
set.seed(123)
samp=sample(1:nrow(tugas),n)
data.train = tugas[samp,]
data.test = tugas[-samp,]
dim(data.train)
dim(data.test)

fit <- rpart(delay~., data = data.train, method = 'class')
summary(fit)
fit$variable.importance

but with fit$variable.importance, i cannot run that as the result is null. how can i fix this?

Answer 1

It doesn't work because all your predictions are the majority class:

fl = https://raw.githubusercontent.com/niharikabalachandra/Logistic-Regression/master/FlightDelays.csv

tugas=read.csv(fl)
dipakai=c(1,2,4,8,10,13) 
l=dim(tugas)[1] 
tugas<-tugas[1:l,dipakai] 

n <- round(nrow(tugas)*0.70)
set.seed(123)
samp=sample(1:nrow(tugas),n)
data.train = tugas[samp,]
data.test = tugas[-samp,]

fit <- rpart(delay~., data = data.train, method = 'class')
table(predict(fit,type="class"))

delayed  ontime 
      0    1541 

You need to solve this issue of imbalanced learning.. Below I just adjust weights to get predictions that are not all majority class, it does not however improve the precision of the model:

wt = ifelse(data.train$delay == "delayed",1.5,1)
fit <- rpart(delay~., data = data.train, method = 'class',weights =wt)
 table(predict(fit,type="class"))

delayed  ontime 
     97    1444

table(predict(fit,data.train,type="class"),data.train$delay)
         
          delayed ontime
  delayed      53     44
  ontime      235   1209

You can get the importance now:

fit$variable.importance
  carrier      dest schedtime   dayweek    origin 
40.275159 23.709600 19.088864 16.221204  9.527087