A pre-averaged pseudo nearest neighbor classifier.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• The k-nearest neighbor (KNN) algorithm is a widely used classification technique.
• KNN's performance degrades with small sample sizes and the presence of outliers.
• Existing methods struggle to effectively handle noisy data in classification tasks.

Purpose of the Study:

• To propose a novel classifier, the pre-averaged pseudo nearest neighbor classifier (PAPNN), to improve classification performance.
• To address the limitations of KNN in small-sized samples with outliers.
• To reduce the negative impact of outliers on classification accuracy.

Main Methods:

• The PAPNN rule involves calculating pre-averaged categorical vectors by averaging pairs of training data points within each class.
• k-pseudo nearest neighbors are identified from these preprocessed vectors for classifying query points.
• The method preprocesses training data to create robust feature representations.

Main Results:

• Extensive experiments were conducted on nineteen numerical and three high-dimensional real datasets.
• PAPNN was compared against twelve other classification methods.
• The proposed PAPNN rule demonstrated effectiveness in classification tasks, particularly for small samples with outliers.

Conclusions:

• The PAPNN classifier effectively improves classification performance in challenging datasets.
• Pre-averaging techniques offer a viable strategy to mitigate outlier influence in KNN.
• PAPNN shows promise for real-world applications requiring robust classification with limited or noisy data.