<p>A new technique enabling to improve feature selection process based on weighted intuitionistic fuzzy (IF) similarity relation (WIFSR) is suggested in the present study. Firstly, we discuss a novel WIFSR by improving the idea of IF similarity relation. Secondly, IF granular structure (IFGS) is established on the basis of WIFSR. Thirdly, IF rough set model is outlined based on the idea of aforesaid IFGS. Next, positive region is computed based on the lower approximation of IF rough set. Then, dependency of decision dimension over set of conditional dimension is calculated based on positive region and cardinality of the decision system. With granular structures, features/dimensions can be represented with different levels of abstraction to provide a dynamic and flexible selection approach. Moreover, we present a WIFSR designed to measure the similarity between features by taking into account their relevancy and non-redundancy. Proposed approach effectively addresses the problem of feature selection by measuring degree of dependency between features in IFGS framework. Mathematical validation is illustrated for all the established notions. Proposed method is experimentally evaluated on various datasets, and we successfully demonstrate its efficiency in terms of determining the inherent features while safeguarding them from later uncertainty and noise. Our experimental results illustrate that the suggested approach, in the context of accuracy and standard deviation, outperforms the existing feature selection methods. At the end, a new scheme is demonstrated to enhance the overall prediction performances of machine learning methods for antiviral peptides.</p>