Anomaly detection with tensorflow models with no labels

Hello everyone,

I am working on a credit card fraud detection project. I want to create a tensorflow model where the last layer is as follows:

tf.keras.layers.Dense(1, activation=“sigmoid”)

But, there is one problem. The data I have consists of only transactoins, their date and time, location, location of credit card registration place and the distance between the two, transaction limit, and the ratio of transaction amount to the card’s credit limit.

I do not know which of the transactions are fraudulent, so I do not have the y (target) variable. When I try to fit the model I have created, it shows me an error.

Is it possible to create a model like the one I have in mind? If not, what are some alternatives?