Bagging Ensemble Classifiers for Imbalanced Big Data

The patent divides the large imbalanced dataset into multiple smaller chunks, where each chunk contains a representative subset of minority cases and majority cases. Multiple base models are trained on different chunks, and their predictions are aggregated through voting. This segmentation allows training on personal devices with limited memory while maintaining classification accuracy through ensemble aggregation.

The patent ensures that each training chunk has a different composition of minority and majority cases, creating local variations in data quality across chunks. Some chunks may have higher concentrations of minority cases while others have more majority cases. This local quality variation allows the ensemble to capture different aspects of the imbalanced data distribution, improving minority class recognition without requiring the entire large dataset to be loaded into memory simultaneously.

The patent trains multiple base models on partial subsets (chunks) of the complete dataset rather than requiring all data to be processed simultaneously. Each base model is trained on a manageable chunk that fits within personal device memory constraints. The ensemble of multiple partially-trained models achieves performance comparable to a single model trained on the complete dataset, enabling cost-effective training on personal computing devices.