Year: 2024
Type: preprint
Abstract: Data normalization is a critical step in machine learning workflows, particularly for clustering algorithms sensitive to feature scaling. This paper introduces a modified Robust Scaling method defined as X1= X−median(X) / mean(X), which combines the resilience of the median with the global sensitivity of the mean. The proposed scaling method is evaluated on standard datasets, including Iris and Wine, using K-means clustering. Results demonstrate that X1 scaling effectively handles outliers, enhances clustering accuracy. Additionally, we provide a detailed analysis of sorted label accuracy, confusion matrices, and the optimal random initialization seed for clustering. The findings highlight the potential of X1 scaling as a robust alternative for preprocessing in machine learning tasks.
Source: Preprints.org
Authors: Maradana Durga Venkata Prasad, Srikanth T
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4404475643