Heart Sound Analysis using Machine Learning

---

• Understand Mel Frequency Cepstral Coefficents (MFCCs)
• Understand k-Nearest Neighbors (kNN) algorithm
• Understand Convolutional Neural Network (ConvNet/CNN)
• Classify abnormal/normal heart sound using MFCCs, kNN, CNN

Dataset

---

Dataset is retrieved from the training set of 2016 PhysioNet/CinC Challenge. There are 3,240 heart sounds colected in uncontrolled environment, saved in .wav format, each lasts from 5s to 120s.

Database	Abnormal	Normal	Total
Training-a	292	117	409
Training-b	104	386	490
Training-c	104	7	31
Training-d	28	27	55
Training-e	183	1958	2141
Total	665	2575	3240

Table: Summary of dataset

Libraries

---

Implementation is in Python3.

• Librosa for reading and extracting MFCCs
• sklearn for kNN
• Keras for CNN

Documents

This project includes:

1. 01 report
2. 01 notebook