Thyroid Disease Prediction: Leveraging Machine Learning for Early Recurrence Detection

Endocrine disorder thyroid disease affects many patients who need precise and prompt diagnostic methods for suitable treatment strategies. This research investigates machine learning model predictability for diagnosing thyroid disease using medical data collected from patients. The pre-processed dataset incorporated patient demographics alongside hormonal levels TSH T3 and T4 as well as lifestyle elements and medical documentation after performing data cleaning operation and encoding alongside feature scaling processes followed by exploratory data analysis. The evaluation of machine learning models used accuracy, precision, recall, and F1 score metrics to test Logistic Regression, Decision Trees, Random Forest and CatBoost. Research results showed that Random Forest together with CatBoost obtained 97.30% accuracy whereas baseline models including Logistic Regression reached only 91.89% accuracy and Decision Tree reached only 94.59 % accuracy. The accuracy of CatBoost received a 1.41% boost through hyperparameter tuning which increased its generalisability capabilities. The error analysis showed that Logistic Regression performed poorly with non-ideal patterns because it produced excessive wrong negative predictions, yet Decision Trees demonstrated issues when dealing with marginally correct inputs. The Random Forest and CatBoost algorithms surpassed other models because they perfectly identified multiple feature relationships. The research shows that ensemble learning methods demonstrate potential to strengthen thyroid disease diagnosis by cutting down misdiagnosis rates while boosting clinical decision support. This research emphasizes that healthcare organizations should establish AI predictive modelling approaches for early healthcare identification alongside personalized treatment protocols which lead to better thyroid disorder diagnosis accuracy.