Attachment recognition in school-age children through multimodal analysis of verbal and non-verbal behaviour

Alsofyani, Huda (2024) Attachment recognition in school-age children through multimodal analysis of verbal and non-verbal behaviour. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 2024AlsofyaniPhD.pdf] PDF
Download (2MB)


Attachment is a psychological construct concerning the affectional and emotional bonds between children and their caregivers. Attachment styles can be one of two main types: Secure attachment which describes a state where the child’s emotional needs are fulfilled, and Insecure attachment when these needs are unfulfilled. The formation of these attachment styles during childhood shapes the internal representation of close relationships which in turn impacts the quality of adulthood relationships and life. Moreover, past studies found that insecure attachment is linked to major issues such as heart diseases and antisocial behaviours. Early psychological interventions can mitigate the potential negative consequences of insecure attachment. To this end, it is essential to identify insecure children as early as possible. Various attachment assessment tests have been devised for infants and children which rely on observing their behaviours when exposed to distress situations. However, these tests suffer from a major drawback that impedes their applicability for population large-scale screenings, which is the need for trained professionals. One promising solution to overcome this challenge is by automating the assessment process and therefore increasing the applicability of the assessment tests.

In this thesis, automatic attachment recognition approaches based on the Manchester Child Attachment Story Task (MCAST) assessment test are proposed based on three main behavioural channels: facial expressions, paralanguage, and language. In addition, this thesis explores the benefit of combining different modalities to enhance the recognition rate. The results show that the attachment styles can be detected automatically with an accuracy of up to 75% by combining paralanguage and language modalities, and an F1-score of up to 68% by combining face and language modalities. Additionally, age and gender based performance analyses are conducted revealing that older children are more likely to express their condition through facial expressions while younger children are more likely to express it through paralanguage and language. Gender based effects are observed too and it is found that female children are more likely to express their condition through facial expressions, while male children tend to express it through language. These findings can enhance the recognition rate because it might be useful to adopt different modalities for different age or gender groups. It is also shown in this thesis that incorporating a confidence measure can increase the applicability of the approaches, for instance, by setting an acceptance level corresponding to 80% accuracy, which can be considered as human-level performance, 77% of the predictions can be accepted using an approach based on combining all of the three behavioural channels.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Colleges/Schools: College of Science and Engineering > School of Computing Science
Supervisor's Name: Vinciarelli, Professor Alessandro
Date of Award: 2024
Depositing User: Theses Team
Unique ID: glathesis:2024-84392
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 25 Jun 2024 09:04
Last Modified: 25 Jun 2024 09:04
Thesis DOI: 10.5525/gla.thesis.84392
Related URLs:

Actions (login required)

View Item View Item


Downloads per month over past year