<p>The increase in yoga practice at home, spurred by digital accessibility and recent health crises, has intensified the demand for automated pose correction systems to compensate for the lack of human guidance. While prior works have effectively recognized and classified yoga poses, the crucial aspect of pose correction remains inadequately addressed. Most existing approaches evaluate all yoga postures using a uniform set of features and fixed thresholds, without accounting for the fact that the relevance of specific joints or angles varies across poses. Furthermore, defining pose-specific thresholds typically requires domain expertise, making such systems less adaptive and more difficult to generalize. Motivated by these gaps, this paper proposes a data-driven framework for automated yoga pose correction based on pose-specific feature relevance. Relevant angular features are systematically ranked using SHAP-based interpretability analysis, and an optimized subset of discriminative features is derived for each pose to enable accurate pose comparison. In addition, this research introduces a systematic multi-strategy optimization mechanism for realigning misaligned joints, considering the combinations of deviations of multiple joints to attain desired angles without violating anatomical constraints. Moreover, the framework provides corrective feedback through multiple visual refinements, adjusting the level of correction based on the quality of the joint deviation after relocation. In addition, the system supports multimodal feedback mechanisms, integrating visual guidance, textual instructions, and speech-based cues to enhance user understanding and engagement.</p>