R 分类树问题

Question

I'm relativly new to machine learning, but I want to create a classification tree based on a world bank dataset. The classification tree must meet 3 characteristics: continent=Europe , currency=euro , income group = high income

This graphic should look like this:

I already tried this, but it does not give me the output that I want:

library(tidyverse)
library(rpart)

WDICountry <- read.csv("https://gigamove.rz.rwth-aachen.de/d/id/pUKMStHbu9orYo/dd/100")

tree1 <- WDICountry %>%
  mutate(europe = ifelse(`2-alpha code` %in% european_countries, TRUE, FALSE),
         euro = ifelse(`Currency Unit` == "Euro", TRUE, FALSE),
         income = as.factor(ifelse(`Income Group` == "  High income", "High income", "non-high")))%>%
  mutate(`Income Group` = as.factor(`Income Group`))%>%
  select(`Income Group`, europe, euro)%>%
  filter(complete.cases(.))%>%
  rpart(data = .,formula = `Income Group` ~ europe+ euro)

plot(tree1)
text(tree1)

Can someone help me?

You can download this dataframe as a csv file here: https://gigamove.rz.rwth-aachen.de/d/id/pUKMStHbu9orYo?10&id=pUKMStHbu9orYo

Answer 1

A decision tree is a model to classify, so if there is insufficient evidence to support the splitting of the variable in classification, the split is not performed. Hence when you plot, not all your variables or possible splits are used.

Below I will tweak some of the parameters to get all the splits to happen, but note this is most likely not the way to construct a decision tree model.

First, to get the data. Note you don't need a ifelse to set a boolean, I have a lot of problems with the variable names, so below is something with the corrected column names, using your csv file:

library(rpart)
library(rpart.plot)
library(dplyr)

WDICountry = read.csv("WDICountry.csv",stringsAsFactors=FALSE)

european_countries = WDICountry[grep("Europe",WDICountry$Region),"X2.alpha.code"]

dat = WDICountry%>%
mutate( europe = X2.alpha.code  %in% european_countries,
       euro= Currency.Unit =="Euro",
       income=as.factor(ifelse(Income.Group =="High income","High income","non-high")))%>%
select(income,europe,euro)%>%
filter(complete.cases(.))

Before fitting the model, see the smallest split:

table(dat$europe,dat$euro)
       
        FALSE TRUE
  FALSE   203    2
  TRUE     35   23

So you need to set the minimum split at the lowest to ensure it splits, and also we set the complexity parameter such that the split proceeds:

mdl = rpart(income ~ europe+euro,data = dat,minsplit=2,method="class",cp=-1)
rpart.plot(mdl)