Improved binary dragonfly optimization algorithm and wavelet packet based non-linear features for infant cry...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Infant cry signals provide crucial information for early diagnosis of conditions like hunger, pain, asphyxia, deafness, jaundice, and prematurity.
• Automated analysis of infant cries can significantly aid in timely medical intervention.

Purpose of the Study:

• To propose a hybrid method combining wavelet packet transform features with an Improved Binary Dragonfly Optimization (IBDFO) algorithm for accurate infant cry signal classification.
• To evaluate the effectiveness of the proposed feature selection and classification approach for diagnosing various infant states.

Main Methods:

• Extracted 568 features from cry signals using wavelet packet transform, Linear Predictive Coding (LPC), and Mel-frequency Cepstral Coefficients (MFCCs).
• Employed the IBDFO algorithm to select the most informative features, addressing the curse of dimensionality.
• Utilized an Extreme Learning Machine (ELM) kernel classifier for binary and multi-class cry signal classification.

Main Results:

• Achieved high accuracies in binary classification, including 100% for asphyxia vs. normal and deaf vs. normal.
• Demonstrated excellent performance in multi-class classification, reaching 100% accuracy for three classes (normal, asphyxia, deaf) and 97.62% for seven classes.
• The IBDFO-selected features (204 out of 568) were sufficient for high-accuracy classification.

Conclusions:

• The proposed method effectively classifies infant cry signals, offering a promising tool for early diagnosis.
• The combination of advanced feature extraction and optimized feature selection enhances classification accuracy for detecting subtle cry signal changes.