Positive‐versus‐Negative Classification for Model Aggregation in Predictive Data Mining

The process of constructing several base models that are then combined into a single classification model for prediction is called model aggregation or ensemble classification. Positive‐versus‐negative (pVn) classification is a new method for the implementation of base models for aggregation. pVn classification involves the decomposition of a k‐class prediction task into m (m < k) subproblems. One base model is constructed for each subproblem to predict a subset of the k classes. The base models are then combined into one aggregate model for prediction. This paper reports studies that were conducted to demonstrate the performance of pVn classification when large volumes of data are available for modeling as is commonly the case in data mining. It is demonstrated in this paper that pVn modeling provides the capability to use a large amount of available data (in a large data set) for base model training. It is also demonstrated that pVn models created from large data sets provide a higher level of predictive performance compared to single k‐class models.