Adaptive KNN Classifier Boosts Accuracy with Granular Ball Computing.

The k-Nearest Neighbor (KNN) algorithm is a widely used classification method, but its performance heavily relies on the appropriate selection of the 'k' value. A fixed 'k' often struggles with varying data distributions, leading to suboptimal accuracy or sensitivity to noise. Researchers have introduced an adaptive KNN approach leveraging granular-ball computing. This method operates in two stages: first, a training stage coarsely partitions the dataset into multi-granularity granular balls, using the Fisher criterion to guide splitting and stopping. This reduces data complexity within each ball. In the prediction stage, the algorithm locates the nearest granular ball to a test sample using a weighted distance. It then constructs an adaptive neighborhood around the sample, dynamically determining the effective 'k' value based on the number of samples within this stable local group. This approach enhances robustness against noise and local perturbations, demonstrating superior accuracy and efficiency compared to existing KNN variants.