Alean Stacked Ensemble Model(lsem) to Enhance the Effectiveness of Classifying Data With Huge Imbalance

Abstract

Knowledge discovery and analysis has become one of the major needs of the current information rich world. Effective information identification and prediction requires effective models. Several machine learning models are available for prediction. This paper concentrates on classification, a supervised machine learning model. An effective classifier can enable effective predictions. However, not all input data are perfect to enable highly accurate classification. Several factors such as data imbalance, noise and borderline entries affect the classifiers. This paper proposes a Lean SVM based Ensemble Model (LSEM) that enables effective classification of data without the need for pre-processing. A heterogeneous ensemble is created using Random Forest and One-Class SVM. The requirement of partial training data for SVM makes the model lean, enabling faster training. Experiment is conducted on data with varied imbalance levels and it is identified that the proposed LSEM operates better than state-of-the-art models and ensembles and hence enabling better predictions.