Developing Thyroid Disorder Prediction Model Using Machine Learning Approach

Abstract

Thyroid disorder is a prevalent disorder affecting endocrine system on a globally, placing a significant burden on healthcare budgets, particularly in low-income countries. Early prediction and prevention of this disorder based on symptoms represent a critical challenge for healthcare sectors, especially in developing nations like Ethiopia. Machine learning has a vital role in the analysis of vast medical datasets and offers solutions to the intricate task of early disease classification. Recently, predicting thyroid disorder prediction has gained significant importance as a task. Despite the current methods for diagnosing this condition, it often involve binary classification, utilize limited datasets, imbalacnce dataset and lack proper validation. Current efforts mostly concentrate on refining models, neglecting feature engineering. this study introduces an approach that addresses these limitations by delving into feature selection for machine learning models. Various techniques are employed, such as FFS, BFE, bi-directional feature elimination, and feature selection through machine learning, involving the use of extra tree classifiers. The Proposed approach aims to predict different types of thyroid disorders: Negative, Hyperthyroidism and Hypothyroidism. The data for this study was gathered from Tikur Anbessa Specialized Hospital and Saint Paul’s Millennium Medical College. the five algorithms employed in the study were Random Forest, Logistic Regression, Support Vector Machine, Adaptive Boosting and Extreme Gradient Boosting. The models were assessed using a stratified 10-fold cross-validation technique. and an analysis of their classification performance, enabling a comparison between the different models. the performance of the given classification techniques was evaluated using accuracy, precision, recall and F1- score. the results of the performance evaluation showed that the XGBoost with the feature selection with cross-validation method performs better than other models, which obtained an accuracy of 98.9 and F1-score of 99.1. the proposed thyroid predictive model classifiers a common thyroid disorder based on a multi-class prediction approach to help domain experts