Please use this identifier to cite or link to this item:
Author: Μπαλκούδη, Μιχαέλα
Title: Machine learning-based mood classification via standardized questionnaires
Date Issued: 2023
Department: Πρόγραμμα Μεταπτυχιακών Σπουδών στην Τεχνητή Νοημοσύνη και Αναλυτική Δεδομένων
Supervisor: Χρήστου Βαρσακέλης, Δημήτριος
Abstract: In recent years, there has been an increase in awareness of mental health issues and it is widely accepted that their early detection is essential to preventing social consequences. The use of questionnaires is a common medical technique for promptly detecting mental health concerns. Some scientists have proposed further automating the diagnosis of one mental condition by utilizing a questionnaire that diagnoses another condition. However, more research and study are required in order to prove the effectiveness of this further automation of the diagnosis of mental disorders and make it practical. This thesis investigates two questions. First whether a standardized memory questionnaire known as the PRMQ (Prospective and Recall Memory Questionnaire) along with a few demographic and general health-related questions, may be used to diagnose depression. Second, we try to investigate the reverse, that is whether memory- related disorders may be diagnosed in patients by using a common questionnaire that makes a diagnosis of depression called the ZUNG Depression Questionnaire (SDS), coupled with the same demographic questions and health-related questions used in the first investigation. The selection of these two mental illnesses is not arbitrary; rather, it is based on their usual co- occurrence and the link that has been found between them. Both questions will be inves- tigating via machine learning techniques. More specifically, question is approached in two ways: as a regression and as a classification task. For each such task, suitable machine learn- ing models are applied and compared in order to find the one with the best performance. The memory-related classification task will turn out to be an imbalanced classification problem, hence appropriate methods, such as resampling during training and cost-sensitive algorithms, are used to resolve it. Our results show that we can diagnose depression through the memory questionnaire, coupled with some demographic questions and health-related questions with an accuracy of approximately 79%. The diagnosis of memory-related issues via the Zung depres- sion questionnaire could not be achieved with adequate accuracy. This does not necessarily imply that we can not diagnose memory-related issues from a depression questionnaire, but more research is needed to improve performance.
Keywords: Depression
Memory issues
Information: Διπλωματική εργασία--Πανεπιστήμιο Μακεδονίας, Θεσσαλονίκη, 2023.
Rights: CC0 1.0 Παγκόσμια
Appears in Collections:ΠΜΣ στην Τεχνητή Νοημοσύνη και Αναλυτική Δεδομένων (Μ)

Files in This Item:
File Description SizeFormat 
BalkoudiMichaelaMsc2023.pdf1.2 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons