This is a new method to the family of fuzzy k-nearest neighbor (FKNN) classifiers based on the use of power means in the calculation of multi-local means that are used in classification of samples. This method is called multi-local power means fuzzy k-nearest neighbor classifier (MLPM-FKNN). The MLPM-FKNN can be adapted to the context (of different data sets), due to the power mean being parametric and thus allowing for testing to find the parameter value that can be optimized for the classification accuracy.
The function of the mlpm-fknn algorithm (mlpm_fknn.m) and Power mean computation (pmean.m) are included. In addition to those files, an example (example_mlpm_fknn.m) of the use of mlpm-fknn classifier is also presented. pmean.m is needed to compute Power mean vectors of the set of nearest neighbor in each class.
Created by Mahinda Mailagaha Kumbure & Pasi Luukka 01/2019 based on Keller's definition of the fuzzy k-nearest neighbor algorithm.
This is the updated version of the original MLPM-FKNN classifier according to the study:
Kumbure, M. M., Lohrmann, C., Luukka, P.: A Study on Relevant Features for Intraday S&P 500 Prediction Using a Hybrid Feature Selection Approach. International Conference on Machine Learning, Optimization, and Data Science (LOD - 2021), Grasmere, Lake District, England – UK (2021).
In this method, training data is grouped into each class first, and the set of k-nearest neighbors for the query sample is searched from each class. Next, multi-local power means vector for each set of nearest neighbors from each class is computed. The rest of the steps are the same as in the original MLPM-FKNN method.