Machine learning helps us flag or predict fraud based on historical data. The most common method for fraud-detection is classification. For a classification problem, a set of data is mapped to a subset based on the category it belongs to. The training set helps to determine to which subset a dataset belongs. These subsets are often known as classes:

In cases of fraudulent transactions, the classification between legitimate and non-legitimate transactions is determined by the following parameters:

• The amount of the transaction
• The merchant where the transaction is made
• The location where the transaction is made
• The time of the transaction
• Whether this was an in-person or online transaction