Towards Automated Emotion Classification of Atypically and Typically Developing Infants

Area of Science:

• Developmental Pediatrics
• Machine Learning in Healthcare
• Infant Neurodevelopment

Background:

• 15 million preterm births annually pose risks for developmental delays.
• Cerebral palsy and other issues impact neuromotor and cognitive development.
• Understanding emotion and movement roles is crucial for diagnosing atypical infants.

Purpose of the Study:

• To evaluate automated emotion assessment for classifying infant development.
• To compare a custom machine learning algorithm with a neural network for infant emotion classification.
• To explore the potential of facial features in automated infant emotion analysis.

Main Methods:

• Developed a custom supervised machine learning algorithm using facial features.
• Compared the custom algorithm's performance against a state-of-the-art neural network.
• Utilized individual and grouped facial features for infant emotion classification.

Main Results:

• A custom algorithm required only three concavity features for effective classification.
• The custom algorithm demonstrated performance comparable to the neural network.
• Facial feature analysis shows promise for automated infant emotion assessment.

Conclusions:

• Automated emotion assessment, particularly using facial features, can aid in classifying infant development.
• Further research will integrate emotion and movement data for predicting neurodevelopmental delays.
• This approach may enhance early diagnosis of conditions like cerebral palsy.