Εφαρμογή τεχνικών μηχανικής μάθησης σε δημογραφικά/κοινωνικο-οικονομικά δεδομένα

Ακριβούλης, Αριστείδης

Application of machine learning techniques in demographic/social-economic data

Master Thesis

Author

Ακριβούλης, Αριστείδης

Date

2026-08

Abstract

This thesis investigates cognitive functioning in older adults and its relationship with demographic, health, and social characteristics through the combined application of multivariate statistical methods and machine learning techniques. Data from the Survey of Health, Ageing and Retirement in Europe (SHARE) and its specialized cognitive substudy, SHARE-HCAP, were used for the analysis. The main objectives were to investigate the multivariate structure of the cognitive tests, examine differences according to cognitive status, and identify broader cognitive, health, and sociodemographic profiles of the participants. Initially, descriptive and inferential analyses were conducted, with appropriate control for the multiplicity of statistical tests. Principal Component Analysis and exploratory Factor Analysis using the principal axis method with oblimin rotation were applied to investigate the structure of cognitive functioning. The resulting factor scores were subsequently combined with demographic, health, and social indicators and used in Cluster Analysis employing the Ward and k-means methods. Prior to clustering, the variables were standardized and thematic weighting was applied so that the different domains of information were represented in a balanced manner. The robustness of the findings was further examined through sensitivity analyses and bootstrap resampling procedures, in which the main stages of the analytical process were re-estimated. Principal Component Analysis revealed a dominant dimension of general cognitive performance and a supplementary dimension that mainly differentiated visuospatial and constructional performance from word recall and recognition. Factor Analysis identified distinct but correlated cognitive dimensions related to word recall and recognition, narrative memory, visuospatial and constructional abilities combined with nonverbal reasoning, as well as attention, mental processing speed, and verbal fluency. The findings therefore support a multidimensional organization of cognitive functioning alongside the presence of a broader common dimension of cognitive performance. Cluster Analysis identified broader participant profiles that differed simultaneously in cognitive, health, and social characteristics. The more favorable profile was associated with better cognitive performance, younger age, higher educational attainment, more favorable health and functional indicators, and a stronger social network. In contrast, the less favorable profile was characterized by lower cognitive performance, older age, lower educational attainment, and less favorable health and functional characteristics, while a greater presence of participants with cognitive impairment was observed within this profile. Despite the clear interpretative differentiation between the profiles, measures of internal separation indicated substantial overlap between them. Nevertheless, the k-means solution demonstrated high sampling stability in the resampling analyses. Overall, the findings highlight the usefulness of combining dimension-reduction techniques, methods for investigating latent cognitive structures, and unsupervised clustering for the description of complex profiles in older age. They also emphasize that cognitive functioning should be examined in conjunction with individuals’ broader demographic, health, and social context. Due to the cross-sectional design, the complete-case analysis, and the absence of sampling weights, the findings apply to the specific analytical sample and should not be interpreted as causal or diagnostic, nor as direct estimates of population-level differences.

Postgraduate Studies Programme

Εφαρμοσμένη Στατιστική

Department

Σχολή Χρηματοοικονομικής και Στατιστικής. Τμήμα Στατιστικής και Ασφαλιστικής Επιστήμης

Number of pages

400

Language

Greek

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/19744

Collections

Τμήμα Στατιστικής και Ασφαλιστικής Επιστήμης

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα