<p>In the domain of imbalanced classification problems, accurately classifying the minority class is a significant area of focus within the machine learning community. The conventional Extreme Learning Machine (ELM) method often exhibits a bias towards the majority class as a direct consequence of the class imbalance. To mitigate this issue, several extensions of ELM have been developed. One such example is the Minimum Variance-Embedded Class-Specific Kernelized ELM (MVCSKELM). Kernelized ELM (KELM) offers better generalization capabilities compared to conventional ELM. The Universum SVM (USVM) addresses class imbalance by incorporating prior information into the classification model by integrating Universum samples to the training set. Numerous other modifications of SVM have been proposed, incorporating Universum samples in the model generation. An ELM-based classification model builds two symmetric planes, one for each class. In contrast, the Universum-based ELM classification model seeks to build a third plane between the two symmetric planes using Universum samples. This paper introduces a novel hybrid framework called Universum-based Graph Embedded class-specific KELM (UGECSKELM), which combines Universum learning with MVCSKELM for the first time to leverage the benefits of both methods. Universum samples are training samples from the same domain that do not belong to any of the target classes. The proposed UGECSKELM maintains comparable computational complexity to MVCSKELM. Experimental results on real-world datasets demonstrate the superiority of our method over current approaches.</p>