Project Motivation and Description:
Stemming from my master’s research work, after finding the S1 and S2 hearts sounds, I wanted to determine whether someone had a heart valvular failure. After reading an academic paper and using an open-source dataset and , I implemented several networks—a BiLSTM, a LSTM, a BiGRU, and a GRU network—to best determine which can classify someone with a heart abnormality or not.
Research Details and Results:
While Support Vector Machines (SVM) performed the best amongst the machine learning algorithms, BiLSTM performed extremely well with an accuracy of 98%.
Most Challenging Part of the Project:
Although the results for the BiLSTM network are promising, I would like to run some tests. The dataset I found online was limited, so the accuracy might be extremely high because of overfitting. In addition, I am trying to integrate the datasets I used for my master’s research to for my designed model.