Weighted Nearest Centroid Neighbourhood
A novel binary classifier based on nearest centroid neighbours is presented. The proposed method uses the well known idea behind the classic k-Nearest Neighbours (k-NN) algorithm: one point is similar to others that are close to it. The new proposal relies on an alternative way of computing neighbourhoods that is better suited to the distribution of data by considering that a more distant neighbour must have less influence than a closer one. The relative importance of any neighbour in a neighbourhood is estimated using the SoftMax function on the implicit distance. Experiments are carried out on both simulated and real data sets. The proposed method outperforms alternatives, providing a promising new research line.
KeywordsNearest Neighbours Classification Nearest Centroid Neighbourhood Parameter selection Similarity measure
Research supported by grant from the Spanish Ministry of Economy and Competitiveness, under the Retos-Colaboración program: SABERMED (Ref: RTC-2017-6253-1); Retos-Investigación program: MODAS-IN (Ref: RTI2018-094269-B-I00); and the support of NVIDIA Corporation with the donation of the Titan V GPU.
- 4.Dua, D., Graff, C.: UCI Machine Learning Repository (2017). http://archive.ics.uci.edu/ml
- 8.Hulett, C., Hall, A., Qu, G.: Dynamic selection of k nearest neighbors in instance-based learning. In: 2012 IEEE 13th International Conference on Information Reuse & Integration (IRI), pp. 85–92. IEEE (2012)Google Scholar
- 9.Jaiswal, S., Bhadouria, S., Sahoo, A.: KNN model selection using modified cuckoo search algorithm. In: 2015 International Conference on Cognitive Computing and Information Processing (CCIP), pp. 1–5. IEEE (2015)Google Scholar