The 2-NN Rule for More Accurate NN Risk Estimation.

Area of Science:

• Machine Learning
• Statistical Pattern Recognition

Background:

• Nearest-neighbor (NN) risk estimation is sensitive to sample size.
• Existing NN rules can be affected by variations in data quantity.

Purpose of the Study:

• To investigate methods for reducing sample size effects in NN risk estimation.
• To enhance the accuracy and stability of NN risk estimates.

Main Methods:

• Designing a specific nearest-neighbor (NN) rule, termed the 2-NN rule.
• Analyzing the asymptotic properties of the 2-NN rule compared to the 1-NN rule.
• Introducing 2-NN polarization for further risk estimate stabilization.

Main Results:

• The 2-NN rule effectively eliminates first-order sample size effects in two-class problems.
• The 2-NN rule's asymptotic value is half that of the 1-NN rule, improving accuracy.
• The 2-NN approach was successfully extended to M-class and 2k-NN scenarios.

Conclusions:

• The 2-NN rule offers a more accurate and stable alternative to the 1-NN rule for risk estimation.
• Proper design of NN rules can mitigate sample size dependency in machine learning.
• The 2-NN approach provides a robust method for classification tasks with varying sample sizes.