Team name: Ulster Engs Diagnosis of patients with congenital or acquired heart diseases is difficult in developing countries due to the lack of healthcare infrastructures and cardiovascular specialists. The screening protocol using phonocardiogram (PCG) or heart sound auscultation may help in the early detection of heart diseases. The detection of heart murmurs using phonocardiography needs a high level of training. Consequently, this study aims to develop a machine learning algorithm that uses PCG signals to help general practitioners in rural areas to distinguish between normal and abnormal subjects. The PCG signals were filtered using a Butterworth bandpass filter (25-400 Hz). Afterward, the signals were converted to images by the application of Mel-frequency cepstral coefficients (MFCC). In addition, 100 time and frequency features were extracted from the signal. We developed a network that has two branches: the first one is a CNN classifier that accepts MFCC images, and the second one is a Multi-layer Perceptron classifier that accepts demographic data alongside the extracted features. Then, the two branches combine to classify the existence of the murmur in the PCG as a present, absent, or unknown. The model was trained firstly on balanced data from the 2022 challenge data and tested on the 2016 challenge data to classify them as normal or abnormal. The PCG signals that were incorrectly classified or have scores lower than 0.7 were added to the 2022 challenge data to increase the data labeled unknown. Then, the 2022 challenge data and the added data from the 2016 challenge were used to train the model which was evaluated on a blind test dataset. The overall score obtained using the challenge evaluation metric is 1900.329.