COOT algorithm is a recent metaheuristic algorithm that simulates American coot birds when moving in the sea. However, the COOT algorithm like other metaheuristic techniques may be stuck in local regions. In this study, a modified COOT algorithm called (mCOOT) is presented which is based on 2 techniques: Opposition-based Learning (OBL) & Orthogonal Learning to overcome these limitations. Moreover, to test the novel algorithm called mCOOT, we apply it to the dimensionality reduction problem using 9 UCI datasets and compare it with the original algorithm and 3 other ones. Results prove the effectivness and superiority of the proposed algorithm in solving feature selection in terms of classification accuracy and selected features numbers.