An improved local mean-based pseudo nearest neighbor classification using different distance metrics

Local mean-based pseudo nearest neighbor (LMPNN) is an efficient classification with accurate prediction, novel methods, and broad application prospects. This paper proposes an improved LMPNN classification using Euclidean, Cosine, Correlation, Spearman, and Jaccard distance. This classifier tests and compares which of the five effective distance metrics can achieve the lowest average error rate, and studies the universal best distance metric of the LMPNN classification to improve the performance of the improved LMPNN. In order to achieve the above goals, this paper conducted extensive classification error experiments on ten representative datasets from the UCI repository. Experimental results show that Cosine can achieve higher prediction accuracy than the Euclidean often used in standard LMPNN.