Machine Learning Based Feature Selection for Early Detection of Thyroid Disorders in Nigeria

• Peter S. IDOKO

  Author

• Iyinoluwa T. IDOWU

  Author

• Emmanuel O. AYODELE

  Author

• Temitope F. SHOLANKE

  Author

• Peter A. IDOWU

  Author

Disorders of the thyroid are regarded as one of the main concerns related to global public health issues. They cause considerable harm in underdeveloped countries like Nigeria. This paper attempts to find the most accurate predictors of Nigerian thyroid disease via using the machine learning approach. Finding relevant features for building the models with robust forecasting reliability is a crucial stage of the machine learning process. All redundant variables should be eliminated from the initial data set for the sake of making the process of model training more effective and avoiding possible cases of overfitting. That is why this paper aims at using machine learning methods for selecting those features suitable for predicting the early development of the disease in the Nigerian population. Clinical indicators (TSH, T3, T4, autoantibodies), demographic parameters (sex, age, body mass index), ultrasound characteristics, and environmental variables (exposure to goitrogens and iodine content) are taken into account. Both a filtering approach and the usage of Random Forest algorithm are utilized to select the best features. As shown by results, Random Forest and Gradient Boosting performed equally well, while Random Forest has slightly better predictive power. Using the entire set of features, Random Forest reached the accuracy of 0.9978, a precision of 0.9986, a recall of 0.9971, F1-score of 0.9978, and an ROC-AUC equal to 0.9999. Gradient Boosting demonstrated the same performance: accuracy = 0.9971, ROC-AUC = 0.9999.

• FJET_21_57_57

Copyright (c) 2026 FUDMA Journal of Engineering and Technology

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.