Enhancing Predictive Accuracy for Differentiated Thyroid Cancer (DTC) Recurrence Through Advanced Data Mining Techniques

Thyroid cancer is becoming more common, and its 20% recurrence rate of which almost half are discovered more than five years after surgery, highlights how difficult it is to distinguish between a true disease relapse and chronic disease brought on by insufficient initial treatment. This ambiguity highlights the complicated dynamics that drive the mortality rates in patients with thyroid cancer. The purpose of this study is to be refining these predictions to control Differentiated Thyroid Cancer recurrence and minimize the risk of recurrence. The dataset was obtained by monitoring a total of 383 patients with 17 attributes. This study adopted a data mining modelling strategy to evaluate the performance, classification accuracy, and cluster distribution, utilizing the Orange data mining software. The Exploratory Data Analysis was conducted to pinpoint the most significant contributors. Subsequently, a variety of supervised techniques were applied to assess the precision of both single and ensemble models in classification. For cluster determination, we implemented several unsupervised learning techniques, including k-means, hierarchical, and Louvain Clustering. The result shows that ensemble stacking algorithm demonstrated superior performance and classification accuracy, achieving impressive scores of 0.971. The analysis of clustering methods, notably k-means and hierarchical clustering, suggested that the dataset could be segmented into two distinct clusters. The most dominant factors in influencing the recurrence of thyroid cancer with strong correlation revealed 'Response', 'Risk', 'Adenopathy', and 'N'. The refinement of the diagnostic model, through the identification of accurate models and key factors, enhances the prediction of Differentiated Thyroid Cancer recurrence.