Automatic Health Speech Prediction System Using Support Vector Machine

In the past couple of years, seeking for automatic disease prediction system (ADPS) on the Internet has been attempted by many users online especially after COVID-19 pandemic. This points to a new generation of medical treatment, especially when the number of internet users is growing day by day. Therefore, automatic disease prediction online applications have gained the attention of many researchers around the world. Through this work, an automatic disease prediction system depending on speech has been designed and implemented. The system aims to predict the type of disease, and the patient is suffering from depending on his voice, but this was not applicable through experiment, so the diseases were divided in to three groups, and the diagnoses were implemented accordingly. The benchmark of this work was the medical speech, transcription, and intent dataset. The features utilized in this work are the smoothness, mel-frequency cepstral coefficient (MFCC), and the spectral centroid variability (SCV) features that proved their high representation to human being medical situation in this work. The noise reduction forward-backward filter was used to remove noise from the wave files recorded online for the high noise noticed in the dataset deployed. A hybrid feature selection algorithm was designed and implemented for this work which combined the output of the genetic algorithm (GA) with the inputs of the neural network (NN) algorithm. Support vector machine (SVM), neural network, and Gaussian mixture model were utilized for classification. The highest results gained according to our design groups are 94.55 and 50.1% according to each disease, both with respect to SVM.