Data loading and preprocessing code for the paper titled 'Fraud Detection with NLP', which demonstrates the use of natural language processing techniques to detect fraud in financial transactions.

This code requires the following Python packages:

• pandas
• numpy
• sklearn
• ast
• tensorflow

These can be installed using pip or another package manager.

To use this code, you will need to have the following files in your working directory:

• Fraud Detection with Natural Language Processing.pkl: a pickled Pandas DataFrame containing the credit card transaction data
• vocab.csv: a CSV file containing a list of action names and their corresponding IDs

Once these files are in place, you can run the code and generate several new columns in the DataFrame, including:

• Action time mean: the average time between actions in a transaction
• Action time std: the standard deviation of the time between actions
• log(amount): the natural logarithm of the transaction amount
• Transaction Type: a string indicating whether the transaction is fraudulent or not
• time_to_first_action: the time between the start of the transaction and the first action taken
• actions_str: a string containing the names of all actions taken in the transaction
• total_time_to_transaction: the total time elapsed from the start of the transaction to its completion

• The code assumes that the actions column in the DataFrame is a list of lists, where each inner list contains a sequence of actions taken in a single transaction.
• The code also assumes that the times column in the DataFrame is a list of lists, where each inner list contains the timestamp (in milliseconds) of each action taken in a single transaction.