In situ and network-based transaction classifying systems and methods
Various embodiments herein each include at least one of systems, methods, and software for in situ and network-based transaction classification. Such embodiments use advanced data analytics and machine learning techniques of consumer's transaction attributes to reduce shrink at checkout. One embodiment, in the form of a method, includes processing a dataset of transactions to identify normal transaction patterns and processing a dataset of transactions that included known fraud to identify variation patterns between the identified normal transaction patterns and the data of each transaction. The method further includes generating at least one pattern model based on the identified normal transaction patterns and the identified variation patterns. In such embodiments, each pattern model typically includes classification values for determining a likelihood of fraud in transactions. The method continues by applying the model to a current transaction to calculate a score indicative of a likelihood of fraud and outputs the score.
1. A method comprising:
processing, on a computer processor, a dataset of transactions to identify normal transaction patterns;
processing, on the computer processor, a dataset of transactions that included known fraud to identify variation patterns between the identified normal transaction patterns and the data of each transaction;
generating, on the computer processor, at least one pattern model based on the identified normal transaction patterns and the identified variation patterns, each pattern model including classification values for determining a likelihood of fraud in transactions;
applying, by the computer processor, the model to a current transaction to calculate a score indicative of a likelihood of fraud; and
outputting, from the computer processor, the score.
2. The method of claim 1 , wherein data of a transaction within the datasets of transactions includes a plurality of data items including data identifying at least one product ID, a date and time, and a point of sale terminal identifier.
3. The method of claim 2 , wherein the data of a transaction includes additional data including data identifying, directly or indirectly, customer data.
4. The method of claim 3 , wherein the customer data includes payment type and amount data, demographic data, purchase history data.
5. The method of claim 1 , wherein applying the model to a current transaction includes comparing current transaction data to patterns represented in the model to obtain a score indicative of a likelihood of fraud being present in the current transaction.
6. The method of claim 5 , wherein:
a score is obtained from the current transaction data with regard to at least some of the patterns represented in the model based on the classification values of the respective patterns; and
potential fraud is indicated when a sum of the pattern scores is greater than a sum threshold.
7. The method of claim 6 , wherein potential fraud is also indicated when an individual pattern score or a sum of a specific group of pattern scores is greater than a threshold for the respective individual pattern or the specific group of patterns.
8. The method of claim 6 , wherein outputting the score includes outputting an indicator of potential fraud.
9. The method of claim 8 , further comprising:
combining the score with a score received or derived from data received from another process; and
when the combined score is greater than a threshold, outputting an intervention interrupt to a point-of-sale terminal where the transaction is being conducted.
10. The method of claim 1 , wherein the dataset of transaction processed to identify normal transaction patterns is a dataset of transactions of a specific store location for which the model will be utilized.
11. A method comprising:
generating, on a computer processor, a model from patterns identified in normal and fraudulent transaction data, each pattern including a classification value that contributes to a transaction classification value indicating a likelihood of fraud in a given transaction;
applying, by the computer processor, the model to a current transaction to obtain classification values to calculate a transaction classification value indicating a likelihood of fraud in the current transaction; and
outputting, from the computer processor, the score as an indicator of potential fraud.
12. The method of claim 11 , wherein the normal and fraudulent transaction data includes a plurality of data items including data identifying at least one product ID, a date and time, and a point of sale terminal identifier.
13. The method of claim 12 , wherein the data of a transaction includes additional data including customer data, the customer data including at least two of payment type and amount data, demographic data, purchase history data.
14. The method of claim 11 , wherein applying the model to a current transaction includes comparing current transaction data to patterns represented in the model to obtain a score indicative of a likelihood of fraud being present in the current transaction.
15. The method of claim 14 , wherein:
a score is obtained from the current transaction data with regard to at least some of the patterns represented in the model based on the classification values of the respective patterns; and
potential fraud is indicated when:
a sum of the pattern scores is greater than a sum threshold; or
an individual pattern score or a sum of a specific group of pattern scores is greater than a threshold for the respective individual pattern or the specific group of patterns.
16. The method of claim 15 , further comprising:
combining the score with a score received or derived from data received from another process; and
when the combined score is greater than a threshold, outputting an intervention interrupt to a point-of-sale terminal where the transaction is being conducted.
17. The method of claim 11 , wherein the normal and fraudulent transaction data the is processed to generate the model of patterns is a dataset of transactions of a specific store or a set of stores where the model will be utilized.
18. A point-of-sale (POS) system comprising:
a computer processor; and
a memory device storing instructions executable by the processor to perform data processing activities comprising:
applying, by the computer processor, a model to a current transaction to obtain classification values to calculate a transaction classification value indicating a likelihood of fraud in the current transaction, the model including data representations of patterns in normal and fraudulent transaction data, each pattern including a classification value that contributes to a transaction classification value indicating a likelihood of fraud in a given transaction; and
interrupting, by the computer processor, operation of the POS system when a transaction classification value exceeds a threshold indicating a likelihood of fraud.
19. The POS system of claim 18 , further comprising:
a network interface device; and
wherein applying the model to a current transaction includes:
transmitting current transaction data via the network interface device to a service to receive a fraud determination in the form of a classification value; and
receiving a response from the service via the network interface device including the classification value.
20. The POS System of claim 18 , wherein the data processing activities further comprising:
combining the classification value with a score received or derived from data received from another process; and
when the combined score is greater than a threshold, interrupting operation of the POS system.