An Efficient Classification of Neonates Cry Using Extreme Gradient Boosting-Assisted Grouped-Support-Vector Network

Area of Science:

• Neonatal care
• Infant communication
• Machine learning in healthcare

Background:

• Infant cries are a primary communication method for newborns.
• Distinguishing cry types (hunger, sleep, discomfort) can be challenging for caretakers.
• Accurate cry interpretation is crucial for timely and appropriate infant care.

Purpose of the Study:

• To develop and validate a novel method for classifying newborn infant cries into three distinct categories: hunger, sleep, and discomfort.
• To enhance caretaker understanding of infant needs through automated cry analysis.
• To improve the efficiency and accuracy of neonatal distress detection.

Main Methods:

• Acoustic feature engineering was employed to extract twelve features from cry signals.
• Random Forests were utilized for variable selection to identify highly discriminative features.
• An extreme gradient boosting-powered grouped-support-vector network was deployed for cry classification.

Main Results:

• The proposed method effectively classified neonate cries into hunger, sleep, and discomfort groups.
• The system achieved a mean accuracy of approximately 91% across various scenarios.
• The developed model demonstrated a rapid recognition rate, identifying emotional cries within 27 seconds.

Conclusions:

• The extreme gradient boosting-powered grouped-support-vector network shows significant potential for accurate neonate cry classification.
• The study provides a reliable tool to assist caretakers in interpreting infant vocalizations.
• The findings highlight the efficacy of advanced machine learning techniques in understanding infant emotional states.