摘要 |
A method and system for making financial or medical decisions. The method comprises training sets of models using classification training with sets of data derived from segregated data sources. Overall weighting of each model within the sets of models are determined for each of the sub-datasets. The sets of models, the overall weighting of each model and a number of examples provided from the data for each of the datasets are transmitted to a central server over a communication network, wherein the central server is configured to determine the relative weights of each of the sets of models in the overall ensemble model based on the number of examples, combine the sets of models, receive new application data, and predict at least one of outcome variables, an uncertainty factor for the variables, and drivers of the outcome variables based on the new application data. |
主权项 |
1. A computer implemented method for determining the likelihood of a certain financial, insurance or health outcome, the method comprising:
receiving at one or more modeling servers, data from underwriting databases of one or more underwriting parties, the data including information from existing life insurance policies; generating by the modeling servers, for each of the plurality of underwriting parties, at least one dataset and sub-datasets of the at least one dataset from the data; training, via the one or more modeling servers, sets of models of a given learning technique for each of the sub-datasets; determining by the modeling servers, overall weighting of each model within the sets of models for each of the sub-datasets; transmitting to a prediction server over a communication network, the sets of models, the overall weighting of each models and a number of examples provided from the data for each of the datasets, the prediction server configured to determine relative weights of each of the sets of models based on the number of examples, combine the sets of models into an ensemble model, receive new application data from underwriting servers, and predict at least one of outcome variables, an uncertainty factor for the variables, and drivers of the outcome variables based on the new application data. |