Cardiovascular diseases (CVDs) stand as the primary reason for worldwide deaths. The phonocardiogram (PCG) analysis provides an inexpensive screening tool which uses non-invasive methods yet automated systems encounter two main obstacles from background interference, sensor precision issues and unbalanced patient data distribution that makes it hard to detect rare cardiac abnormalities. The research develops and tests an original deep learning system which solves these problems. The pre-processing stage of this method divides PCG signals into cardiac cycles before using wavelet-based denoising to enhance Signal-to-Noise Ratio (SNR) values. The system generates a new hybrid feature map through vertical stacking of multiple feature sets which includes Mel-spectrograms and MFCCs for sensor distortion resistance and wavelet and statistical features for additive noise resistance.