<![CDATA[The classification of baby cries using machine learning is essential for developing automated systems that can assist caregivers in identifying and responding to the needs of infants promptly and accurately. This study aims to improve upon previous research relating to the Cry Baby Dataset, which has highly imbalanced data. We combine oversampling and undersampling techniques using SMOTE and ENN, along with data augmentation through pitch shifting and noise addition to address the data imbalance issue. The processed data was then modeled using Convolutional Neural Networks (CNN). The study yielded an overall accuracy of 88%, with balanced accuracy observed across all classes, effectively mitigating data imbalance. This represents a notable advancement compared to previous research, which often encountered challenges with unbalanced accuracies across classes. The classes identified include recordings of baby cries attributed to belly pain caused by colic, recordings related to burping, recordings associated with discomfort or other symptoms, recordings of hungry baby cries, and recordings indicating fatigue or the need for sleep. This shows a significant improvement from previous studies, which had very unbalanced accuracy for each class.]]>
Copyrights © 2024
