Data imbalance poses a significant challenge in machine learning, especially in binary classification t asks, a s traditional algorithms tend to favor the majority class, often resulting in biased models. This paper proposes two novel clustering-based techniques: Pseudo-Cluster Classification and Pseudo-Cluster Two-Step Classification, designed to improve balance and model generalization for the minority class. By partitioning the majority class into distinct sub-clusters, these methods enhance class representation in the dataset, facilitating improved accuracy. Experimental results indicate that these methods outperform default classification approaches, particularly in terms of Macro F1-Score and MCC.

This work aims to illuminate age-related variations in feature importance for predicting the future performance of professional soccer players. Through an in-depth analysis of the FIFA dataset, encompassing a wide array of player attributes and performance metrics including team-related factors, notable trends across distinct age groups have been discerned.