An Improved Pseudo Nearest Neighbor: Minkowski Distance-weighted Classification Algorithm

The k-Nearest Neighbors (k-NN) algorithm is a widely recognized non-parametric classification method known for its simplicity and effectiveness. Nonetheless, this method is constrained by issues like equal-weighted voting and sensitivity to irrelevant features. This paper presents an improved algorithm titled Distance-weighted Pseudo Nearest Neighbor (PNN) with Minkowski distance. The Distance-weighted PNN merges distance weighting and PNN to bolster classification precision by allocating weights to neighbors according to their distances from the query point. Neighbors in closer proximity receive greater weights, addressing imbalanced datasets and mitigating the influence of noisy or irrelevant features. The inclusion of the Minkowski distance metric, which encompasses metrics such as Euclidean distance and Manhattan distance, provides adaptability in discerning underlying data patterns. The tunable parameter within the Minkowski distance offers a means to mold the metric, tailoring it to specific data nuances. Comprehensive trials on benchmark datasets underline the enhanced performance of the presented algorithm in comparison to traditional k-NN and other distance-weighted k-NN variants. Specifically, the Distance-weighted PNN with Minkowski distance continually attains superior classification outcomes, especially in settings marked by imbalanced class distributions and the presence of noisy attributes.