System and method for scalable cost-sensitive learning

Abstract

A method (and structure) for processing an inductive learning model for a dataset of examples, includes dividing the dataset into N subsets of data and developing an estimated learning model for the dataset by developing a learning model for a first subset of the N subsets.

Description

BACKGROUND OF THE INVENTION [0001] 1. Field of the Invention [0002] The present invention generally relates to a technique of inductive learning. More specifically, an inductive model is built both “accurately” and “efficiently” by dividing a database of examples into N disjoint subsets of data, and a learning model (base classifier), including a prediction of accuracy, is sequentially developed for each subset and integrated into an evolving aggregate (ensemble) learning model for the entire database. The aggregate model is incrementally updated by each completed subset model. The prediction of accuracy provides a quantitative measure upon which to judge the benefit of continuing processing for remaining subsets in the database or to terminate at an intermediate stage. [0003] 2. Description of the Related Art [0004] Modeling is a technique to learn a model from a set of given examples of the form {(x1, y1), (x2, y2), . . . , (xn, yn)}. Each example (xi, yi) is a feature vector, xi....

AI Technical Summary

Benefits of technology

[0019] In view of the foregoing exemplary problems, drawbacks, and disadvantages of the conventional methods, an exemplary feature of the present invention is to provide a structure and method for an inductive learning technique that significantly increases the accuracy of the basic inductive learning model.

[0020] It is another exemplary feature of the present invention to provide a technique in which throughput is increased by at least ten to twenty times the throughput of the basic inductive learning model.

[0026] In a sixth exemplary aspect of the present invention, also described herein is a method of at least one of increasing a speed of development of a learning model for a dataset of examples and increasing an accuracy of the learning model, including dividing the dataset into N subsets of data and developing an estimated learning model for the dataset by developing a learning model for a first subset of the N subsets.

[0028] With the above and other exemplary aspects, the present invention provides a method to improve learning model development by increasing accuracy of the ensemble, by decreasing time to develop a sufficiently accurate ensemble, and by providing quantitative measures by which a user (e.g., one developing the model or implementing an application based on the model) can decide when to terminate the model development because the ensemble is predicted as being sufficiently accurate.

Problems solved by technology

Because there is a cost associated with the letters, and each individual will either donate different amount of money or does not donate at all, this model is cost-sensitive.

Fraud challenging and investigation are not free.

There is an intrinsic cost associated with each fraud case investigation.

As should be apparent, there is also a second cost associated with the development of the model that is related to the cost of the computer time and resources necessary to develop a model over a database, particularly in scenarios where the database contains a large amount of data.

A problem recognized by the present inventors is that, in current learning model methods, the entire database must be evaluated before the effects of the hypothetical parameters for the test model are known.

Depending upon the size of the database, each such test scenario will require much computer time (sometimes many hours or even days) and cost, and it can become prohibitive to spend so much effort in the development of an optimal model to perform the intended task.

Hence, there is currently no method that efficiently models the cost-benefit tradeoff short of taking time and computer resources to analyze the entire database and predicting the accuracy of the model for whose parameters are undergoing evaluation.

Images