Toward Early Detection of Neonatal Birth Asphyxia Utilizing Ensemble Machine Learning Approach

Perinatal asphyxia is one of the top three causes of neonatal death in developing countries, killing over 1.2 million newborns yearly. Asphyxia cannot be definitively diagnosed early on visually or physically; instead, it can only be diagnosed medically. In this research, an ensemble machine learning-based approach is proposed to detect infant asphyxia at the early stage. Mel-Frequency Cepstral Coefficients (MFCCs), which divide each feature’s values into the time domain and the frequency domain, were originally used in the technique to evaluate feature extraction methodologies. Unwanted noise, outliers, missing numbers, label encoding, and other difficulties are eliminated using pre-processing techniques. By applying the random oversampling (ROS) method, data balance is achieved. After analyzing and evaluating the performance of the proposed model, it is observed that the highest accuracy 99.29% is obtained using the combination of logistic regression and K-nearest neighbor with a 0.007% rate of error.