N3 and BNN: Two New Similarity Based Classification Methods in Comparison with Other Classifiers

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• K-nearest neighbors (KNN) is a foundational classification algorithm.
• Object pairwise similarities are key to many classification tasks.
• Evaluating new methods against established ones is crucial for progress.

Purpose of the Study:

• To introduce and evaluate two novel classification methods: N3 (N-nearest neighbors) and BNN (binned nearest neighbors).
• To compare the performance of N3 and BNN against nine established classification techniques.
• To assess the efficiency and effectiveness of the proposed methods across diverse datasets.

Main Methods:

• Development of N3 (N-nearest neighbors) and BNN (binned nearest neighbors) algorithms.
• Comparative analysis using 32 diverse literature datasets.
• Performance evaluation against nine well-known classification methods, including KNN and SVM/RBF.

Main Results:

• N3 exhibited the highest average efficiency, performing comparably to Support Vector Machine with Radial Basis Function kernel (SVM/RBF).
• BNN demonstrated superior performance compared to the traditional K-nearest neighbors (KNN) method on average.
• Both novel methods showed robust performance across datasets varying in size, dimensionality, and class distribution.

Conclusions:

• N3 represents a highly efficient classification method, offering a competitive alternative to SVM/RBF.
• BNN provides an improved approach over standard KNN, enhancing classification accuracy.
• The proposed N3 and BNN methods offer valuable contributions to the field of machine learning classification.