Artificial Intelligence Meets Endocrinology - A Machine Learning-Based Approach to Thyroid Disease Diagnosis Using Feature Selection Methods - Model results and decision-support framework

Aug 3, 2026

Thyroid disease develops when the thyroid gland grows abnormally or fails to produce enough hormones, leading to serious health consequences. Early, efficient identification is essential for better clinical intervention and ongoing disease management. This research integrates advanced machine learning with feature selection strategies to improve thyroid disease classification using a preprocessed dataset from UCI. Chi-Square and Recursive Feature Elimination select optimal features, while Linear Discriminant Analysis supports dimensionality reduction. Models including MLP, Gradient Boost, and RNN are evaluated with accuracy, precision, recall, and F1-score, with Gradient Boost reaching 99% accuracy.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

Thyroid function and hormonal regulation

Classification of thyroid disorders

Risk factors and prevalence

Methods (Feature selection and modeling)

Feature selection: Chi-Square and Recursive Feature Elimination

Dimensionality reduction: Linear Discriminant Analysis (LDA)

Classification models and evaluation metrics

Artificial Intelligence Meets Endocrinology: A Machine Learning-Based Approach to Thyroid Disease Diagnosis Using Feature Selection Methods Aftab Ahmad Khan1, Bakhtiar Khan1, Muhammad Arif1, Waseel ud Din2, Wahab Khan1, Yasir Tayyab Khayyam4 Ashraf Ullah1, Kalim Ullah3 1Department of Computer Science, University of Science & Technology Bannu, Pakistan 2Department of Computer Science, Birmingham City University, United Kingdom 3Department of Electrical Engineering, University of Science & Technology Bannu, Pakistan 4Gomal Research Institute of Computing (GRIC), Faculty of Computing, Gomal University, D.I. Khan, K.P.K, Pakistan * Correspondence: Aftab Ahmad Khan; Email: [aftabaak7@gmail.com](aftabaak7@gmail.com) Citation | Khan. A. A., Khan. B, Arif. M, Din. W, Khan. W, Khayyam. Y. T, Ullah. A, Ullah. K,“Artificial Intelligence Meets Endocrinology: A Machine Learning-Based Approach to Thyroid Disease Diagnosis Using Feature Selection Methods”, IJIST, Vol. 07 Issue. 04 pp 2383-2398, October 2025 Received| August 31, 2025 Revised| October 08, 2025 Accepted| October 10, 2025 Published| October 12, 2025. Thyroid Disease (TD) arises when the thyroid gland either grows abnormally or does not generate enough thyroid hormones, and might cause serious health issues and consequences. Early and efficient identification of thyroid disease is important for improved clinical intervention and disease management. By combining sophisticated and advanced machine learning models with a range of advanced feature selection strategies, this research study aims to enhance the classification of thyroid disease based on a machine learning based diagnostic system. The preprocessed dataset used in this study and the trials were taken from the machine learning repository at the University of California, Irvine (UCI) . We employ two popular feature selection techniques- Chi-Square, and Recursive Feature Elimination, and a dimensionality reduction technique Linear Discriminant Analysis (LDA), and to choose the best features from the dataset for experiments. After selecting the most suitable features, they were then used to train and test the machine learning models: MultiLayer Perceptron (MLP), Gradient Boost (GB), and Recurrent Neural Network (RNN) . Evaluation matrices, accuracy, precision, recall, and F1-score were used to assess models'performance. The experimental results show that the machine learning model Gradient Boost (GB) outperformed the other models and yielded an accuracy of 99%, indicating its ability to classify the Thyroid Disease (TD) accurately. The proposed research work helps to create an intelligent decision-support system for medical diagnostics by offering an understandable and reliable framework for Thyroid Detection. Keywords: Thyroid Disease, Hormones, Feature Selection, Linear Discriminant Analysis, Chi-Square, Recursive Feature Elimination, Machine Learning Based Diagnostic System October 2025 |Vol 07 | Issue 04 Page |2383 Introduction: The thyroid, one of the body’s most vital organs, produces hormones essential for numerous physiological processes. Once released into the bloodstream, these hormones circulate throughout the body, regulating growth and metabolism. This indicates that the thyroid directly influences how the body utilizes energy and sustains overall physiological balance. The thyroid gland is perfectly positioned to perform its functions because it is located in the neck, just below the Adam's apple. Understanding thyroid function is essential, as it serves as a key basis for investigating and diagnosing various disorders linked to hormonal imbalances. By producing hormones necessary for growth and metabolism, the thyroid ensures that the body's energy needs are met, promoting overall health and well-being [1] . Thyroid disease is ranked 2nd after diabetes by the prominent healthcare organization WHO (World Health Organization) . Iodine-deficient areas are home to almost one-third of the w

FAQs

What problem does the study address in thyroid disease diagnosis?

The study targets improved identification of thyroid disease by building a machine-learning diagnostic system that classifies thyroid conditions more accurately, supporting clinical decision-making.

Which feature selection and dimensionality reduction methods are used?

Chi-Square and Recursive Feature Elimination are used for feature selection, and Linear Discriminant Analysis (LDA) is applied for dimensionality reduction before modeling.

How are the machine learning models evaluated and what was the best result?

Models are assessed using accuracy, precision, recall, and F1-score. Gradient Boost (GB) outperforms the other models and achieves 99% accuracy for thyroid disease classification.

Aug 3, 2026

Upload by Clementine