I am new to R and I am trying to play around with the data from here. I try to oversampling it but the Error in model.frame.default happen.

1. The first trial

oversample_data <- ovun.sample(class ~ ., data = sample_dataset, p = 0.5, seed = 1, method="over")$data

But

Error in model.frame.default(formula = class ~ step type amount : object is not a matrix

is shown.

2. That's why I come up with second trial and turn the dataset into matrix first and then do oversampling

org_dataset <- as.matrix(org_dataset[complete.cases(org_dataset), ])

data_balanced_over <- ovun.sample(class ~ ., data = org_dataset, p = 0.5, seed = 1, method = "over")$data

But it says

Error in model.frame.default(formula = class ~ step type amount : 'data' must be a data.frame, not a matrix or an array

It makes me so confused... What is the right way to do oversampling?

CodePudding user response：

The problem is the formula you're setting for ovun.sample. There is no variable named class in the dataset you're referring to. The documentation of the ROSE package for the formula says that

The left- hand-side (response) should be a vector specifying the class labels. The right- hand-side should be a series of vectors with the predictors.

Thus, you'll have to specify a variable holding the class labels. Given the dataset, I assume this would be isFraud. The call, then would be

oversample_data <- ovun.sample(isFraud ~ ., data = sample_dataset, p = 0.5, seed = 1, method="over")$data