A Machine-Learning Approach for Children’s Pain Assessments Using Prosodic and Spectral Acoustic Features

This paper presents a comprehensive study on the application of Artificial Intelligence (AI) techniques for classifying the reasons behind an infant’s cry. The aim of the study is to develop an effective model that can accurately categorize infants’ cries using Machine Learning (ML) and Deep Learning (DL) classification methods. The research focuses on two distinct datasets, namely DonateACry and BabyCry, in order to explore the best approach for cry classification. The study includes a preprocessing step with data augmentation techniques to address dataset imbalances. Two feature extraction methods are employed: audio spectrograms and a hybrid approach combining prosodic and spectral acoustic features. Different classification models, including K-Nearest Neighbors (K-NN), Dense Artificial Neural Networks (ANNs), and Convolutional Neural Networks (CNNs), are evaluated using the f1 score and accuracy. The results demonstrate that the hybrid approach of prosodic and spectral acoustic features outperforms audio spectrograms, pro-viding a more comprehensive representation of infants’ cries and improving classification accuracy. The research showcases the potential of AI-driven advancements in revolutionizing childcare and parenting practices, contributing to improved well-being and care for infants. Overall, this study demonstrates the effectiveness of AI techniques in understanding infant communication and emphasizes their potential to improve care for infants.