Back to Top

Paper Title

ANALYZING THE EFFECTS OF DATA IMBALANCE ON THE PERFORMANCE OF NEURAL NETWORKS IN MULTI-CLASS CLASSIFICATION TASKS

Authors

Keywords

  • data imbalance
  • neural networks
  • multi-class classification
  • oversampling
  • undersampling
  • cost-sensitive learning
  • data augmentation
  • model performance
  • precision-recall
  • machine learning

Article Type

Research Article

Issue

Volume : 1 | Issue : 2 | Page No : 1-9

Published On

August, 2024

Downloads

Abstract

This paper investigates the effects of data imbalance on the performance of neural networks in multi-class classification tasks. Data imbalance, where certain classes are underrepresented, poses significant challenges in training effective models, often leading to biased predictions and reduced overall accuracy. The study explores various strategies to mitigate these effects, including oversampling, undersampling, cost-sensitive learning, and data augmentation. By conducting experiments on real-world datasets, this research provides a comparative analysis of how these techniques influence the performance metrics such as accuracy, precision, and recall. The findings highlight the critical role of addressing data imbalance in enhancing the reliability of neural networks, particularly in scenarios involving complex multi-class classification tasks. The results suggest that while certain techniques offer substantial improvements, there remain challenges that warrant further investigation. This paper contributes to the ongoing discourse on optimizing neural networks for imbalanced data scenarios, offering practical insights for researchers and practitioners alike.

View more >>

Uploded Document Preview