Data analytics model selection through champion challenger mechanism
A method for forecasting includes obtaining input data from a data store, using a processor to forecast future data with a currently selected model, detecting a trigger event using a processor, training alternative models in a model family or from multiple families on the input data based on detecting in response to detecting the trigger event, identifying a replacement model from the alternative models using a processor, and using a processor to forecast future data with the replacement model.
1. An information handling system comprising:
a memory including a data store; and
a model challenger circuit to communicate with the memory, the model challenger circuit to:
obtain input data from the data store using a processor that accesses a computer memory of the data store to obtain continuously accumulated data as the input data, the input data including customer survey data, customer purchase history, customer browse history, customer transaction categories, or any combination thereof;
use a currently selected model to forecast future purchasing patterns;
detect a trigger event using the processor, the trigger event including a scale of fluctuations in the data input, a change in a customer purchase pattern including unexpected purchases, altered purchase timing, or any combination thereof;
determine whether the scale of fluctuations of actual or expected value crosses a threshold;
in response to the scale of fluctuations of the actual or the expected values crossing the threshold, detect the trigger event;
in response to the detection of the trigger event, train alternative models based on the input data;
use an intra-family challenge to train and test models to identify an optimal model from a newly selected model family to identify a replacement model, wherein the newly selected model family is from a plurality of model families, the replacement model having a best fit to the input data, wherein the best fit is based on an error rate, an operational reliability over time, or any combination thereof;
forecast future purchasing patterns using the replacement model;
replace the currently selected model with the replacement model such that future purchasing pattern predications are performed using the replacement model; and
provide a product recommendation or targeted promotion to a user based on the future purchasing pattern prediction.
2. The information handling system of claim 1 , wherein the model challenger circuit further to forecast future sales patterns with the replacement model a plurality of times at a fixed interval.
3. The information handling system of claim 2 , wherein the model challenger circuit further to provide a model quality score as a function of time from the plurality of times.
4. The information handling system of claim 1 , wherein the model challenger circuit further to create an individualized sale for the user.
5. The information handling system of claim 1 , wherein the product recommendation or targeted promotion is selected to upsell or cross-sell the user.
6. The information handling system of claim 1 , wherein the model challenger circuit is a processor of the information handling system.
7. The information handling system of claim 1 , wherein the alternative models are located in a model family or in multiple model families.
8. A method for targeted advertising, the method comprising:
obtaining input data from a data store, the input data including customer survey data, customer purchase history, customer browse history, customer transaction categories, or any combination thereof;
using a currently selected model to forecast future purchasing patterns;
detecting a trigger event using a processor that accesses a computer memory of the data store to obtain continuously accumulated data as the input data, the trigger event including a scale of fluctuations in the data input, a change in a customer purchase pattern including unexpected purchases, altered purchase timing, or any combination thereof;
determining whether the scale of fluctuations of actual or expected value crosses a threshold;
in response to the scale of fluctuations of the actual or the expected values crossing the threshold, detecting the trigger event;
in response to the detecting of the trigger event, training alternative models in a model family or from multiple model families on the input data;
using an intra-family challenge to train and test models to identify an optimal model from a newly selected model family to identify a replacement model, wherein the newly selected model family is from a plurality of model families, the replacement model having a best fit to the input data, wherein the best fit is based on an error rate, an operational reliability over time, or any combination thereof;
forecasting future purchasing patterns using the replacement model;
replacing the currently selected model with the replacement model such that future purchasing pattern predications are performed using the replacement model; and
providing a product recommendation or targeted promotion to a user based on the future purchasing pattern prediction.
9. The method of claim 8 , further comprising using the processor to forecast future sales patterns with the replacement model a plurality of times at a fixed interval.
10. The method of claim 9 , further comprising:
providing a model quality score as a function of time from the plurality of times.
11. The method of claim 8 , further comprising:
creating an individualized sale for the user.
12. The method of claim 8 , wherein the product recommendation or targeted promotion is selected to upsell or cross-sell the user.
13. A non-transitory computer-readable medium including code that when executed by a processor causes the processor to perform a method for targeted advertising, the method comprising:
obtaining input data from a data store, the input data including customer survey data, customer purchase history, customer browse history, customer transaction categories, or any combination thereof;
using a currently selected model to forecast future purchasing patterns;
detecting a trigger event by accessing a computer memory of the data store to obtain continuously accumulated data as the input data, the trigger event including a scale of fluctuations in the data input, a change in a customer purchase pattern including unexpected purchases, altered purchase timing, or any combination thereof;
determining whether the scale of fluctuations of actual or expected value crosses a threshold;
in response to the scale of fluctuations of the actual or the expected values crossing the threshold, detecting the trigger event;
in response to the detecting of the trigger event, training alternative models in a model family or from multiple model families on the input data;
using an intra-family challenge to train and test models to identify an optimal model from a newly selected model family to identify a replacement model, wherein the newly selected model family is from a plurality of model families, the replacement model having a best fit to the input data;
forecasting future purchasing patterns using the replacement model;
replacing the current model with the replacement model; and
providing a product recommendation or targeted promotion to a user based on future purchasing pattern prediction;
forecasting future sales patterns with the replacement model a plurality of times at a fixed interval; and
providing a model quality score as a function of time from the plurality of times.
14. The non-transitory computer-readable medium of claim 13 , wherein the best fit is based on an error rate, an operational reliability over time, or any combination thereof.
15. The non-transitory computer-readable medium of claim 13 , further comprising:
creating an individualized sale for the user.
16. The non-transitory computer-readable medium of claim 13 , wherein the product recommendation or targeted promotion is selected to upsell or cross-sell the user.