Method of building predictive models on transactional data

摘要

The present invention is drawn to method of building predictive statistical models based on transactional data utilizing a set of aggregation modules to provide scalar input for logistic regression, neural networks or radial basis functions models. Each transactional source of data is processed by a dedicated aggregation module. The output of aggregation modules can be saved so when the new transactional records become available the output of the model can be updated just by processing these new records only. Parameters of the aggregation module are calculated simultaneously with the parameters of the traditional module during model training.