Nearest Neighbors by Adaptive Simulated Annealing

The nearest neighbour (KNN) supervised classification technique is widely known and used. This technique can be expensive computationally for some applications. In order to improve KNN in relation to the time required for the classification, is proposed an adaptation using Adaptive simulated annealing, a heuristic method inspired by heat treatment, in order to determine similar samples. The modified technique was evaluated with classification problems that are present in the database UCI. The datasets are evaluated in some parameters, these are compared with the results in time and accuracy to explain the behavior of the results. At end is demonstrated that the method reduces the total execution time and its efficiency is comparable with the KNN algorithm based on partitioning trees in datasets with some restrictions.