Impact of Multiplicative noise on Information loss characteristics of datasets
Pratiksha Soori, Shifali Santhosh, V Shashidhar · 2023
The rise of the “big data” notion has sparked a surge in interest in data mining across various disciplines, including business intelligence, scientific discovery, and healthcare. The primary objective is to gather information from large data sets. However, concerns about privacy have caused the rise of privacy-preserving data mining (PPDM), which aims to create data-mining methods without invading privacy. In this paper, original data was modified by applying multiplicative noise on datasets and studying its impact on the information characteristics of the datasets. The extent of dispersion or spread in the attribute values caused by noise was also examined. The multiplicative noise moved in the direction of distorting the difference between the newly added noise column and the original column of values, according to the mean variation error throughout the seed value. The median is more resistant to outliers and noise than the mean is. Although multiplicative noise caused considerable distortion, the median was comparatively constant compared to the mean. Multiplicative noise has alternate frequencies for mode values because it is more dependent on the scaling factors. The standard deviation is increased by scaling the values. Due to this, multiplicative noise also has data loss, which may lead to certain measurements becoming saturated or small-scale changes vanishing.