The newest MFG4.0 dissertation introduces nearest neighbor-based novel soft computing techniques for classification and regression applications
This dissertation focuses on supervised machine learning techniques – classification and regression. In particular, the emphasis is on the fuzzy k-nearest neighbor (FKNN) algorithm that has received substantial attention in classification problems due to its efficacy and flexibility.
In classification, learning from data can be challenging for many algorithms due to uncertainties and inconsistencies in the data. A typical issue associated with most classification problems is that class distributions in the data are imbalanced – meaning data points do not equally represent the classes in a class variable, which can significantly affect classification performance. Given this issue, this research develops solution techniques based on the FKNN algorithm.
The first proposed approach is the multi-local Power mean fuzzy k-nearest neighbor (MLPM-FKNN), which uses class prototype local mean vectors instead of individual training instances in the learning part. The second proposed method is the Bonferroni mean-based fuzzy k-nearest neighbor (BM-FKNN), an enhancement of the MLPM-FKNN method using the Bonferroni mean to compute class prototype local mean vectors. The findings of the tests with different real-world data sets demonstrate the robustness and efficacy of the proposed approaches for class imbalance problems.
Figure: Kumbure, M. M., Luukka, P., & Collan, M. (2020) A new fuzzy k-nearest neighbor classifier based on the Bonferroni mean. Pattern Recognition Letters, 140, 172-178
Mahinda Mailagaha Kumbure
Junior Researcher, LUT University
In addition to the classification context, the rationale underlying the FKNN algorithm for regression problems is successfully generalized, and concurrently, a novel Minkowski distance-based fuzzy k-nearest neighbor (Md-FKNNreg) method is proposed. This method is tested on several different real-world data sets, and the results show that it achieves the best performance compared to the several state-of-the-art regression methods.
Mahinda Mailagaha Kumbure defended his dissertation in the field of Business and Management (Business Analytics) at LUT University on the 10th of November 2022. The title of his doctoral thesis is Novel fuzzy k-nearest neighbor methods for effective classification and regression.
Read more
Mailagaha Kumbure, Mahinda (2022-11-10) Novel fuzzy k-nearest neighbor methods for effective classification and regression
Keywords:
machine learning, classification, regression, feature selection, prediction, class imbalance, fuzzy k-nearest neighbor, local means, performance
