A novel explainable AI-based approach: early-stage Thyroid disease classification

Abstract

As of the most current information, thyroid illnesses remain widespread on a worldwide scale. The rate of hypothyroidism among Americans is around 4.6%, with a higher incidence seen in women and older individuals. Meanwhile, hyperthyroidism affects around 1.3% of the population. Globally, iodine deficiency continues to be a primary factor contributing to thyroid problems, especially in areas with limited availability of iodineenriched diet or supplements. Concerns about thyroid disease should be raised for human health due to the thyroid gland's involvement in regulating human metabolism and its essential impact on overall well-being. Through this research on the thyroid, 3 types of thyroid disease can be identified (hyperthyroid, hypothyroid, negative). To do this research I used several machine learning models (Logistics Regression, Decision Tree, Naive Bayes, Random Forest, Bagging Classifier, XGBoost, SGD, AdaBoost, Grid Search Bagging, Grid Search Logistic, Support Vector Machine (SVM), ANN) etc. Among all the models some of the best performing models are: Bagging accuracy 99.04%, XGBoost accuracy 98.78%, Random Forest accuracy 98.67%, Grid Search Bagging accuracy 98.54%. Among these models, the Bagging classifier model performed best, so here selected this model as the best and final model. Then added Explainable AI to the final model. Explainable AI's job here is to explain how the model is making decisions based on which features to predict results. Previous papers have dealt with 1 or 2 classes. worked on 3 classes (hyperthyroid, hypothyroid, negative) at once to overcome that limitation, So, here can say that the time spent between these tasks is reasonable.