A Quick K-Nearest Neighbor Algorithm with Aggregated Centroids

Pattern classification is a powerful tool for gaining insights into a large data set and predicting which class the unknown data points will fall into. K-nearest neighbors (KNN), as the most fundamental classification algorithm, is applicable in many contexts. However, KNN is computationally expensive because the algorithm needs to loop through every single data point multiple times to compute distance. We propose a new, faster KNN method with aggregated centroids, or Aggregation KNN, that condenses the k nearest neighbors into one single centroid after voting and therefore achieves the purpose of reducing the computation time by reducing the number of distance calculations required. While most algorithms shorten the run time at an expense of accuracy, our experiment on several data sets shows that the AKNN method both reduces the run time and maintains the original KNN's accuracy. Finally, we use generative art to visualize and compare KNN with AKNN.