Discriminative Patterns-based Online Sequential Extreme Learning Machine (DPOS-ELM) for Prediction Framework

Himanshu Shama and Manoj Kumar

In regression and classification prediction models, major goals are interpretability and accuracy. On a set of simple features, less performance and high interpretability are shown by generalized classifiers. From high order interaction between features, detailed set of discriminative patterns can be extracted using proposed Discriminative Patterns-based Online Sequential Extreme Learning Machine (DPOS-ELM) technique. This extraction is used to produce accurate regression and classification. In first stage, large set of high-order patterns are generated by training the samples. In tree based model, prefix paths to leaf nodes from root node are explored as discriminative patterns.In order to fit with generalized linear model, most effective pattern combination is searched by DPOS-ELM. The interpretability and effectiveness of DPOS-ELM are utilized to make a prediction.This component of fast and effective pattern extraction enables strong predictability and interpretability of Discriminative Pattern-based Prediction framework (DPPred). DPOS-ELM provides better accuracy when compared with DDPMine and DPPred for three datasets such as Adult, Hypo and Sick. The results are implemented via the use of Matrix laboratory (MATLAB) with classification metrics such as accuracy and error.

Volume 11 | 11-Special Issue

Pages: 244-251

DOI: 10.5373/JARDCS/V11SP11/20192954