Imbalanced regression refers to methods that focus on predicting rare values in cases where the distribution of the continuous target variable is imbalanced. Compared to imbalanced classification, an additional challenge in imbalanced regression is how to define rare values and distinguish them from regular values, which is essential for accurately predicting rare values in low-density regions. To address this challenge, this paper first proposes a novel rare region identification strategy (KK-means). This method combines kernel density estimation with K-means clustering, allowing for the systematic identification and merging of sparse sample points in the target variable space into continuous rare intervals. Based on KK-means, the paper then introduces a rare value self-boosting imbalanced regression method via rare region identification (RareBoost). RareBoost first extracts information from the identified rare intervals through label density ratio weighting and dynamically adjusts sample weights during bootstrap sampling to enhance the model""s focus on the rare regions, allowing the model to better learn the characteristics of rare value samples during training. Subsequently, a Stacking meta-learner is used to integrate base learners with "rare value awareness" capabilities, forming a rare value prediction model that balances global efficiency and local accuracy. Experimental results demonstrate that RareBoost outperforms traditional methods in key metrics such as ANLL, RMSE, and $R^{2}$. Therefore, RareBoost provides an effective tool for addressing imbalanced regression tasks and demonstrates strong competitiveness in this field.