Methods and arrangements to detect fraudulent transactions

A non-transitory storage medium containing instructions, which when executed by a processor, cause the processor to perform operations, the operations to:

• train a neural network based on purchase histories of multiple customers, the neural network trained to recognize patterns of transactions of the multiple customers;
• associate the neural network with a specific customer of the multiple customers;
• retrain the neural network pretrained with the purchase histories of multiple customers to predict non-fraudulent purchases for the specific customer based on a purchase history of the specific customer, the neural network retrained a plurality of times with sets of transactions specific to the customer to recognize patterns of transactions of the specific customer and predict purchases associated with the specific customer, and wherein each set of transactions includes a number of transactions selected based on one or more rules, wherein the one or more rules comprise a rule to prevent multiple purchases of a particular type within a specified time to be included in a set of the sets of transactions used to retrain the neural network, and a particular type of transaction is based on whether a transaction card is present or not present to perform purchases;
• receive transaction data, the transaction data to describe a purchase made associated with the specific customer;
• perform an inference computation, by the neural network based on the transaction data, to determine a deviation of the purchase from predicted purchases, wherein the determination is based on the transaction data;
• determine whether the deviation of the purchase from the predicted purchases for the specific customer exceeds a deviation threshold;
• and
• cause a message to display on a display device in response to a determination that the deviation of the purchase exceeds the deviation threshold, the message to identify the purchase as outside the predicted purchases predicted by the neural network.

A computer-implemented method, comprising:

• receiving, by a mobile device of a specific customer and from a server, a neural network trained by the server based on purchase histories of multiple customers, the neural network trained to recognize patterns of transactions of the multiple customers;
• retraining, by the mobile device of the specific customer, the neural network pretrained with the purchase histories of multiple customers to predict non-fraudulent purchases for the specific customer, wherein the neural network is retrained a plurality of times with sets of transactions specific to the customer based on a purchase history of the specific customer, and wherein each set of transactions includes a number of transactions selected based on one or more rules, wherein the one or more rules includes a rule to prevent multiple purchases of a particular type within a specified time to be included in a set of the sets of transactions used to retrain the neural network, and a particular type of transaction is based on whether a transaction card is present or not present to perform purchases;
• receiving, by the mobile device, data about a transaction associated with the specific customer;
• determining, by the neural network on the mobile device, a deviation of the transaction from the predicted non-fraudulent purchases based on the data about the transaction;
• comparing, by the mobile device, the deviation of the transaction associated with the specific customer against a deviation threshold;
• generating, by the mobile device, a notification in response to a determination that the deviation of the transaction from the predicted non-fraudulent purchases exceeds the deviation threshold, the notification to identify the transaction;
• presenting, on the mobile device, the notification to a device associated with the specific customer;
• and
• training, by the mobile device, the neural network based on the transaction in response to a determination that the transaction is a fraudulent transaction.