This paper presents analytical techniques to improve redundancy and relevance assessment for precise selection of features in practical multi-class raw datasets. We propose a matrix-rank based $k$-medoids algorithm that guarantees to output all independent medoids. The new algorithm uses matrix rank as a robust indicator, while a traditional $k$-medoids algorithm depends on specific datasets and how the distance between any of two features is defined. Another advantage is that the total number of operations in the nested loops is bounded, different from some $k$-medoids algorithms that involve random search. Sparse regression is an efficient tool for feature relevance analysis, but its outcome can depend on what labeled datasets are employed. A compensation method is introduced in this paper to handle the unequality of class-occurrence in a practical raw dataset. To assess the proposed techniques quantitatively, an existing Industrial Control System (ICS) dataset is used to perform intrusion detection. The numerical results generated from this case study validate the effectiveness and necessity of the proposed analytical framework.