CODEWITHSUNDEEP
rsamplingdownsamplingoversampling
Aether
Aether

Reputation: 1

I am trying to use ROSE to help sampling imbalance. My ovun.sample code is creating empty values, how can I fix this?

I am trying to use ROSE to help with an imbalanced dataset. I am about 90% there, but I am having trouble with my ovun.sample code. When I run the ovun.sample code, it does not create a "over", "under" or "both" dataset, the values are showing in R as NULL (empty), rather than as data. I would appreciate any trouble shooting on how this can be fixed!

set.seed(123)
ind <- sample(2, nrow(credit), replace = TRUE, prob = c(0.7, 0.3))
train <- credit[ind==1,]
test <- credit[ind==2,]


# Data for Developing the Predictive Model
table(train$DEFAULT)
prop.table(table(train$DEFAULT))
summary(train)


# Sample balancing (Over-, Under-, Both)
library(ROSE)
over <- ovun.sample(DEFAULT~., data = train, method = "over", N = 996)$credit
table(over$DEFAULT)

under <- ovun.sample(DEFAULT~., data = train, method = "under", N = 414)$credit
table(under$DEFAULT)

both <- ovun.sample(DEFAULT~., data = train, method = "both",
                    p = 0.5, seed = 213, N = 705)$credit
table(both$DEFAULT)


# Predictive Model (Random Forest)
library (randomForest)
rftrain <- randomForest(DEFAULT~., data = train,
                        ntree = 500, mtry = 10)
rfover <- randomForest(DEFAULT~., data = over,
                       ntree = 500, mtry = 10)
rfunder <- randomForest(DEFAULT~., data = under,
                        ntree = 500, mtry = 10)
rfboth <- randomForest(DEFAULT~., data = both,
                        ntree = 500, mtry = 10)


# Predictive Model Evaluation with test Data
library(caret)
confusionMatrix(predict(rftrain, test), test$DEFAULT, positive = '1')
confusionMatrix(predict(rfover, test), test$DEFAULT, positive = '1')
confusionMatrix(predict(rfunder, test), test$DEFAULT, positive = '1')
confusionMatrix(predict(rfboth, test), test$DEFAULT, positive = '1')```

Upvotes: 0

Views: 1514

Answers (1)

Kruthika
Kruthika

Reputation: 31

I had the same issue. Try changing the code as follows:

over <- ovun.sample(DEFAULT~., data = train, method = "over", N = 996)$data

Similarly change it for the rest. It works.

Upvotes: 1

Related Questions