Aristotle University Of Thessaloniki (AUTH)
Αξιολόγηση λεπτής κινητικότητας με χρήση τεχνικών βαθιάς μηχανικής μάθησης σε παιδιά προσχολικής ηλικίας
Abstract
dc:descriptionΤhis study focused on exploring methods and developing techniques for the automated assessment of fine motor skills and cognitive development in preschool-aged children using deep learning techniques. The research involved 757 children aged 49 to 76 months (4-6 years) from cities and villages within the Regional Unit o fSerres. The children's fine motor development was evaluated using the Griffiths Developmental Scale II, while their cognitive age was assessed with the Goodenough-Harris Draw-A-Person test. The study trained deep learning models such asResNetV2, InceptionV3, MobileNetV2, VGG16, and EfficientNet B0 and B1, and also developed a new model called MotorSkills CNN. The results showed that the reliability of the Griffiths II fine motor scale in categorizing children into the same developmental group as the Goodenough-HarrisDraw-A-Person test was high (93.92%). Although the true positive rate across the six classes was lower compared to other models (31%), the EfficientNet B0 model recorded higher accuracy in categorizing children based on fine motor skills. Eventhough EfficientNet B0 exhibited one of the lowest accuracy rates (35.87%) among the pre-trained networks used, it was considered more effective in categorizing children according to their cognitive development. The new MotorSkills CNN mode lcategorized a total of 124 images (30.46%) in the same category as Griffiths II and127 images (29.48%) in the same category as the Goodenough-Harris Draw-A-Person test. In conclusion, the assessment of the developmental level of the drawings and the measurement of the number of depicted details provide reliable indicators for both the level of fine motor skills and cognitive development in children. In categories with a large volume of images, the models demonstrated acceptable performance, while in categories with a limited number of images, the training and performance of the models were deemed insufficient. This finding serves as a starting point for enriching the dataset and further research using other methods.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Strikas, Konstantinos
- Στρίκας, Κωνσταντίνος
Subjects
dc:subject × 14- Γνωστική ανάπτυξη
- Συνελικτικά νευρωνικά δίκτυα
- Αναπτυξιακή δοκιμασία Griffiths ΝοΙΙ
- Δοκιμασία Goodenough-Harris Draw-A-Person
- Cognitive development
- Convolutional neural networks
- Griffiths developmental scale II
- Goodenough-Harris Draw-A-Person
- Εκπαίδευση
- Κοινωνικές Επιστήμες
- Εκπαίδευση και Έρευνα της εκπαίδευσης
- Education
- Social Sciences
- Education and Educational Research
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57946