Accurately classifying cardiac arrhythmias is a pivotal challenge in biomedical signal processing, crucial for diagnosing and treating heart conditionsry . This study explores the enhancement of 1D Convolutional Neural Network-Gated Recurrent Unit (1D CNN-GRU) models, a cornerstone in processing sequential data, through a hybrid data balancing approach that integrates class weights and resampling techniques. Addressing the challenge of imbalanced datasets, which can severely bias predictions and diminish the performance of models, especially in minority classes, our research presents a comprehensive solution. Utilizing the MIT-BIH arrhythmia dataset, we demonstrate our methodology’s significant impact on model accuracy, achieving a remarkable accuracy of 0.99, sensitivity of 0.93, and specificity of 0.99. These results not only highlight the effectiveness of our hybrid technique in creating a balanced learning environment but also show its superiority in enhancing model performance when compared with the use of standalone class weights or resampling techniques. Our findings underscore the potential of this approach to improve the predictive capabilities of 1D CNN-GRU models across various applications, setting a new benchmark for research in the field of deep learning and offering valuable insights for future advancements.