<p>This paper introduces FusionClass, a hybrid classification framework designed to overcome the common challenges faced by traditional classifiers. Many existing algorithms encounter significant difficulties when dealing with issues such as class imbalance, overlapping categories, and sparse datasets. These limitations often result in poor accuracy and reduced reliability, particularly when the quality or distribution of data is problematic. FusionClass addresses these challenges by integrating three complementary techniques—One-Class SVM, k-Nearest Neighbors, and Classification and Regression Trees (CART)—into a unified framework. Unlike conventional approaches that rely heavily on resampling techniques to balance datasets, FusionClass directly manages these issues without additional preprocessing. To assess its performance, FusionClass was evaluated on several benchmark datasets with varying levels of imbalance and class overlap. The experimental analysis revealed that the proposed framework consistently outperformed widely used classifiers in terms of both predictive accuracy and robustness. Its capacity to handle diverse and complex datasets highlights its adaptability and effectiveness in real-world scenarios. Overall, the findings position FusionClass as a promising advancement in the field of classification, offering a practical and efficient solution for improving model performance when faced with challenging data characteristics.</p>