UNVEILING HIDDEN INSIGHTS: DIMENSIONALITY REDUCTION FOR PROSTATE CANCER DATA WITH PCA AND GAUSSIAN MIXTURE MODEL

Abstract

Applying machine learning techniques for dimensionality reduction of data is a commonly used approach to improve computational efficiency and accuracy of data analysis. Available dimensionality reduction techniques may be categorized into supervised and unsupervised methods. One of the oldest and most popular unsupervised techniques is Principal Component Analysis (PCA), which does not use class labels. It reduces the dimensionality of a dataset while preserving as much variability (i.e., statistical information) as possible. In this study, we apply a multivariate mixture model comprising three components on simulation and real datasets to reduce data and learning volumes. To this end, significant components are first identified by determining the initial values of location parameters using k-means and the default values in the R package. The expectation-maximization algorithm is used to estimate the model parameters. The Bayesian Information Criterion (BIC) is evaluated   for the two sets of initial values, revealing that both approaches  are computationally efficient. Consequently, the proposed technique effectively reduces dimensionality, retains high variance, and achieves reliable clustering. This enhancement increases computational efficiency and improves result quality, making it essential for high-dimensional datasets.

Received: October 17, 2024
Revised: January 13, 2025
Accepted: January 20, 2025