Classifying infant cry patterns by the Genetic Selection of a Fuzzy Model

Infant crying analysis is an important tool for identifying different pathologies at a very early stage of the life of a baby. Being able to perform this task with high accuracy is therefore important and required as a medical support system to assess a baby's health. In this research we propose an automatic classification model for infant crying for early disease detection. Our model mainly consists of two phases: (a) an acoustic features acquisition from the Mel Frequency Cepstral Coefficient and the Linear Predictive Coding from signal processing and (b) the selection/creation of an optimized fuzzy model through the Genetic Selection of a Fuzzy Model (GSFM) algorithm. GSFM searches for the best model by choosing a combination of a feature selection method, a type of fuzzy processing, a learning algorithm together with its associated parameters that best fit the input data. Our approach improves the predictive accuracy on the identification of the cause of crying and clearly helps to differentiate between normal and pathological cry. Experimental results show a significant accuracy improvement when using our optimized genetic selection method for most of the cases.