Advanced Data Mining Techniques for Medicinal and Societal Sciences: A Comprehensive Review

This research presents novel data mining techniques designed to enhance predictive modeling and classification performance, particularly within the fields of social and medical sciences. The study is divided into two parts. Part I addresses the challenge of identifying and ranking key factors influencing population growth. Unlike previous approaches, which often overlook input dependencies, our method integrates decision tree clustering with Cohen’s d index to assess the magnitude of each variable’s effect. Applied to real U.S. city-level data, our method highlights poverty as a significant, previously underreported factor influencing population growth. Part II focuses on improving classification accuracy in imbalanced datasets, where traditional models often fail to detect minority classes effectively. We propose a subsampling-based strategy coupled with support vector machines (SVM) to enhance classification performance. Experiments using 11 publicly available datasets, mainly from healthcare, demonstrate the superiority of our approach over conventional methods, including cost-sensitive and synthetic sampling techniques. These contributions offer robust tools for more accurate decision-making in population studies and medical diagnostics.