Journal of Applied Research in Electrical Engineering

Journal of Applied Research in Electrical Engineering

A Hybrid Technique Combining Robust and DNN for Thyroid Disease Prediction on Imbalanced Dataset

Document Type : Research article

Author
Department of Electrical Engineering, YI.C., Islamic Azad University, Tehran, Iran
Abstract
Thyroid disease poses an important public health risk, reducing the quality of life and increasing healthcare expenditures. In recent years, Artificial Intelligence (AI) algorithms have become essential tools for diagnosing and predicting thyroid disease. Although various AI-based algorithms have been applied to this task, their performance is often hindered by imbalanced data. This research proposes a hybrid approach that combines robust techniques and Deep Neural Networks (DNN) to address this challenge in thyroid disease prediction. The present paper employs the publicly available thyroid disease dataset, obtained from the University of California, Irvine machine learning repository. To manage the data imbalance, Random Under (RU) and Random Over (RO) sampling techniques are applied. A comparative analysis reveals that the hybrid approach, combining robust and DNN with RO sampling, significantly outperforms the RU sampling method on the imbalanced dataset. To assess the effectiveness of the proposed technique, the variant criteria are employed such as accuracy, F1-score, recall, precision, Area Under Curve (AUC), confusion matrix, and Receiver Operating Characteristic (ROC) curve. The hybrid approach achieves superior results, with a testing accuracy of 98.71%, recall of 100%, precision of 97.45%, and an F1-score of 98.71%. These findings demonstrate the robustness and effectiveness of the approach, underscoring its potential for real-time disease prediction.

Highlights

  • Integration of robust scaling and deep neural networks (DNN) for imbalanced thyroid disease prediction
  • Application of Random Over (RO) and Random Under (RU) sampling to address class imbalance
  • Leverage robust scaling to reduce outlier influence and enhance training stability
  • Achieving 98.71% accuracy with RO-sampled hybrid robust DNN architecture
  • Superiority over recent deep learning approaches on imbalanced thyroid datasets

Keywords
Subjects

 
Volume 4, Issue 1
Winter and Spring 2025
January 2025
Pages 1-10

  • Receive Date 21 September 2024
  • Revise Date 06 June 2025
  • Accept Date 01 September 2025