Arrhythmia Classification for Heart Attack Prediction

Introduction Proper classification of heart abnormalities can lead to significant improvements in predictions of heart failures. The variety of patient attributes that factor into arrhythmia classification and the number of resulting arrhythmia classes make this a complex problem to solve. Here, I use UCI’s arrhythmia database[1] and apply a supervised learning algorithm, implemented in Matlab, to train and classify test data into 16 separate classes of heart conditions. The dataset is contains 452 patients and 279 features per patient. The features include patient attributes such as age, height, weight, and gender as well as quantized ECG data such as the amplitude and width of the R, Q, and S waves. The patients are separated into 16 different classes, including one class of normal, 14 classes of various heart conditions (coronary heart disease, AV block, etc.), and one class of other. A small fraction of features are missing in the original dataset as well. I implemented a random forests training algorithm combined with feature selection and data sampling techniques for maximum classification accuracy across all 16 classes. In general, the work presented here used 70% of the samples for training and 30% of the samples for testing.