Multi-Granularity Perceptual Ensemble Learning Model with An Application

For the multi-granularity perceptual problem of structured data and its machine learning tasks, proposing a new ensemble learning framework, multi-granularity perceptual ensemble learning model (GEL), which introduced the concepts of granularity, granule and granular layer. GEL mainly includes multi-granularity perceptual data derivation algorithm (MPDDA), selection and training of base learners, and granule weighted ensemble strategy (GWES). GEL is a supervised learning model, in which the MPDDA provides it with multi-level and perspective on data and to solve problems, this enhanced the interpretability of the model. In GEL, it mainly adopts algorithm of decision tree type as the base to study, GWES will improve the ensemble effect. Finally, a user online behavior abnormal detection task is taken as the experimental application object to prove the effectiveness of GEL model. The experimental results showed that GEL model can improve the prediction effect to a certain extent, and its interpretability is significantly better than that of traditional ensemble learning methods. It can provide decision-makers with a reference of multi-granularity perspective and reduce the cognitive bias and prediction error caused by machine learning model viewing things from a single perspective.