Background: The field of computational surgery involves the use of new technologies to improve surgical safety and patient outcomes. Two open problems in this field include smart surgical tools for identifying tissues via backend sensing, and classifying surgical skill level using laparoscopic tool motion. Prior work in these fields has been impeded by the lack of a dynamic discriminant analysis technique capable of classifying data given systems with overwhelming similarity. Methods: Four new machine learning algorithms were developed (DLS, DPP, RELIEF-RBF, and Intent Vectors). These algorithms were then applied to the open problems within computational surgery. These algorithms are designed with the specific goal of finding regions of data with maximum discriminating information while ignoring regions of similarity or data scarcity. The results of these techniques are contrasted with current machine learning algorithms found in the literature. Results: For the tissue identification problem, results indicate that the proposed DLS algorithm provides better classification than existing methods. For the surgical skill evaluation problem, results indicate that the Intent Vectors approach provides equivalent or better classification accuracy when compared to prior art. Interpretation: The algorithms presented in this work provide a novel approach to the classification of time-series data for systems with overwhelming similarity by focusing on separability maximization while maintaining a tractable training routine and real-time classification for unseen data.