Supervised Machine Learning Algorithms for Arrhythmia Classification and Diagnosis

EasyChair Preprint 3030

Abstract

Preventing arrhythmic risks in patients with cardiac ailment has been a global concern. No reliable method for diagnosing Arrhythmia exist. Machine learning techniques can be used to predict the advent of arrhythmic risk and recommend appropriate measures to ensure patient safety. In this paper, we implement seven machine learning algorithms to predict Arrhythmia. Our model promises to produce better results than the existing VF15 algorithm and cardiologists.
Additionally, our study also advocates that machine learning in arrhythmia identification can reduce diagnos- tic expenses by minimizing type-I and type-II errors. Our study implements a random forest, gradient boosted trees, artificial neural networks, support vector machine, XG-boost, logistic regression, and ensemble method. All the models in this study have been evaluated using classification accuracy, precision, recall, F1 score, gain, lift, ROC, and confusion matrix. A comparative analysis of all the models highlights the strengths and weaknesses of individual algorithms. The ensemble method yields the highest accuracy (0.84), gain (0.31), F1 score (0.81), and ROC (0.87). It produces moderately high false-negative (14%), whereas, support vector machine with an accuracy of 0.71, generates the highest recall value of 0.91 (sensitivity) and a minimum false-negative (11%). The best performing model in this study outperforms the accuracy exiting VF15(62%) technique by more than 20% margin and sensitivity of cardiologist (0.78) by 13%.

Keyphrases: Arrhythmia, Artificial Intelligence, Artificial Neural Network, Support Vector Machine, machine learning

@booklet{EasyChair:3030,
  author    = {Avishek Choudhury and Nikhli Shetty},
  title     = {Supervised Machine Learning Algorithms for Arrhythmia Classification and Diagnosis},
  howpublished = {EasyChair Preprint 3030},
  year      = {EasyChair, 2020}}