This study introduces an innovative rough set model, termed -Adhesion Rough Set Approximation (ARSA), which is developed by employing the concept of adhesion neighborhoods. Building on the classical rough set approximation (RSA) framework, ARSA extends its core concepts to offer a more adaptable and expressive approach to data analysis. The primary motivation behind this model is to address limitations in Pawlak’s original formulation, particularly in handling incomplete or insufficient data. ARSA not only enhances the precision of subset approximations but also establishes a more robust theoretical environment for analyzing complex information systems. Fundamental properties of the proposed model are explored, and interrelationships between various approximation operators are systematically examined. In addition, we improved the Walrus Optimizer algorithm and used it as a feature selection technique that used ARSA as the objective function. To assess the performance of the developed model, a set of experimental series is conducted using UCI datasets and compared with other models. Moreover, the applicability of the developed model is evaluated by using two plant disease datasets related to Tomato and cucumber. The results show the high performance of the developed model in terms of performance metrics.