- author = "Adane Nega Tarekegn",
- title = "Machine Learning for Modeling the Risk of Frailty Syndrome in the Elderly Population",
- school = "University of Turin",
- year = "2020",
- address = "Italy",
- month = oct,
- keywords = "genetic algorithms, genetic programming",
-
URL = "
https://iris.unito.it/handle/2318/2017557",
- size = "228 pages",
-
abstract = "Frailty is a common clinical condition used to
describe older people who are more vulnerable to
stressors and therefore have a higher risk of negative
health outcomes, such as disability, fracture, and
premature death. Several definitions have been proposed
in the literature to conceptualize and operationalize
frailty. However, a universally accepted definition of
frailty is still lacking, making it difficult to
effectively target community services to older adults.
Despite its challenges, frailty is not an irreversible
process and can be reversed or delayed from its
progression. Therefore, it is argued that it should be
detected early.
This thesis focusses on early detection of frailty conditions among older adults in order to provide proactive interventions and, consequently, to maintain well being and quality of life. Most early studies have focused on frailty detection by analyzing the physical performance of individuals, with a relative paucity of an administrative database. However, with the increasing number of the aging population and the growing number of frail elderly, methods to identify frailty within an administrative database are current surveillance priorities. Frailty detection using large administrative databases could capture a complex interplay of a wide variety of heterogeneous factors associated with frailty. Thus, discovering interesting patterns from such large administrative healthcare data is an important application that requires appropriate analytical tools to exploit it fully. Machine learning is a promising tool that is well suited for the analysis and capture of complex patterns within a large dataset.
This thesis presents the application of machine learning as a potential solution for modeling frailty conditions using administrative health database comprising elderly people aged 65 years and above. Both supervised and un-supervised machine learning methods have been explored to develop various models, such as for detecting and predicting adverse outcomes associated with frailty. In supervised learning, both single-label and multi-label classification methods have been examined for building frailty prediction models, while in unsupervised learning, cluster analysis is applied to identify clinically relevant clusters of complex patients. Validation of clustering results and imbalanced data classification are the most difficult problems in the machine learning paradigm. The work presented in this thesis devises new approaches for evaluating the quality of clustering results and proposes a hybrid approach for solving the imbalanced problem in multi-label learning.
Overall, the machine learning models are targeted to assist in the decision-making process aimed at achieving specific clinical health outcomes of the elderly, as well as guide the allocation of healthcare resources and reduced costs.",
- notes = "Supervisors: Giuseppe Costa, Mario Giacobini, Fulvio Ricceri",
