Treffer: Development and Modeling of the K-Nearest Neighbors Algorithm Using Coloured Petri Nets.

The K-nearest neighbor algorithm is among the most widely used methods for classification and regression problems in machine learning and is preferred over many other methods because of its simplicity and availability as a tool in many open-source software libraries. Despite its importance, it is usually offered as an implementation with limited insights into the algorithm steps, which can lead researchers to miss valuable comprehension about the intrinsic details of the method itself. To mitigate this, we introduce a K-Nearest Neighbor (KNN) model implementation that relies on colored Petri nets to enhance the understanding and graphical visualization of the algorithm. The proposed approach uses CPN Tools for modeling the KNN algorithm and conducting tests with various recognized classification and regression datasets from the literature. The model was validated via a comparison with a Python KNN implementation using the Scikit-learn library. An application example of dynamic system modeling is presented in the context of fault detection for robotic manipulators. The results showed that the proposed implementation achieved performance equivalent to the Python implementation and that it allows for a detailed diagnosis, enabling greater understanding and highlighting the relevant steps of the operation of complex discrete event systems. [ABSTRACT FROM AUTHOR]