Rice classification plays a critical role in global agriculture, impacting the food security of over 3.5 billion people worldwide. As the complexity of this task has grown, advanced Machine Learning (ML) and Deep Learning (DL) models have emerged as powerful tools for accurate variety identification. These sophisticated algorithms excel in processing extensive datasets, leveraging key characteristics such as grain size, color, and texture to make precise predictions. However, a persistent challenge in this field is a class imbalance, where certain rice varieties are significantly underrepresented in datasets. This imbalance can severely impact model performance, particularly for minority classes, leading to biased predictions that favor more abundant varieties. Many existing models struggle to effectively address this issue, often prioritizing majority classes at the expense of overall generalization across all rice types. To tackle this critical gap in rice classification, we introduce a novel hybrid model: the XGBoost Multi-Layer 33Perceptron (XGB-MLP). This innovative approach is specifically designed to handle class imbalance, ensuring fair and accurate classification across all rice varieties, regardless of their representation in the dataset. Our model demonstrates remarkable versatility, effectively accommodating binary and multi-class scenarios while maintaining robust performance in the face of imbalanced data. As part of our contribution, we have also developed a new dataset intentionally incorporating class imbalance. This dataset serves as a rigorous benchmark for evaluating our model's performance against existing works in the field. The results of our comprehensive evaluation are compelling, with the XGB-MLP model achieving outstanding accuracy across various classification tasks: 99.86 % for binary class, 99.95 % for multi-class, and 98.46 % for a challenging multi-class scenario using a merged dataset. These impressive results not only surpass the performance of existing systems but also firmly establish the XGB-MLP model's efficacy in diverse rice classification tasks. By effectively addressing the longstanding challenge of class imbalance, our approach represents a significant advancement in the field, offering a more reliable and equitable solution for rice variety identification. This breakthrough has far-reaching implications for agricultural management, food security, and the broader application of machine learning in crop science.