Using ensemble neural network based on sampling for multiclass classification

¹ Faculty of Engineering, Universitas Trunojoyo Madura, East Java, Indonesia, 69162
² Faculty of Agriculture, University of Trunojoyo Madura, East Java, Indonesia, 69162
³ Faculty of Animal Husbandry, Islamic University of Malang, East Java, Indonesia, 65144

^* Corresponding author: bain@trunojoyo.ac.id

Abstract

Multiclass data classification with class imbalance causes classification performance to decrease, especially in the Neural network method. Research shows that the model proposed by eNN can improve model performance for imbalanced data in the selection of superior quality in beef and cattle data. The results of the Ensemble ANN study with adaboost are able to understand complex relationships by measuring the level of correlation with the target class produced. This study aims to overcome the problem of data imbalance in the ensemble neural network method by comparing the oversampling method with undersampling, so that more representative synthetic data is obtained. Performance evaluation is processed using precision, recall and accuracy calculations. Research on superior local Madura cattle data The RUS-eNN method produces the highest average accuracy value compared to others, reaching 98.00% with a recall value of 100%. While the ROS-eNN method produces a difference in accuracy value that is not so far away, namely 97.69%. The research on the sampling-based eNN approach has better accuracy than without using data replication in improving its performance.