Classification of baby cries using machine learning algorithms

People are constantly in communication with each other, and they generally do this through language. Until a newborn baby acquires this skill, crying is the most effective way for them to express themselves. While baby cries may be perceived as disturbing and meaningless by adults, they actually contain a lot of information. When these pieces of information are analyzed using appropriate methods, the reason behind the baby's crying can be revealed. Understanding the cause of the baby's cries is crucial for their health and happiness. In this thesis study, an attempt has been made to interpret baby cries using signal processing techniques and classify them with machine learning algorithms. For this purpose, a dataset containing baby cry audio signals categorized into five classes was used. The dataset was divided into equal parts for each partitioning process, and a separate dataset was created for each part. Then, feature extraction processes were applied to each dataset, and performance values were measured using classification algorithms. These measurement results were examined, and it was observed that higher performance values were achieved on the datasets generated using the data augmentation method. Six algorithms were modeled with the generated datasets, and the highest performance score of 99.51% was obtained with the Extreme Gradient Boosting (XGBoost) algorithm.