Адаптивний алгоритм вибору методу імпутації пропущених значень у задачах інтелектуального аналізу даних

Abstract

In this paper, methods for imputing missing values in numerical datasets are analyzed. The study examines mean, median, and kNN imputation, as well as their impact on the distribution and integrity of the data. An adaptive algorithm is proposed, which allows determining the appropriate imputation method for each feature depending on the distribution shape, presence of outliers, and correlations between features. A real numerical dataset with artificially introduced missing values was used for the experiment. The obtained results confirm the necessity of an adaptive approach to imputation in data analysis and machine learning.