An Automated Machine Learning System for Generating Myocardial Infarction Location Classifiers Using Lead-I ECG Signal
Journal
Proceedings - 2023 14th IIAI International Congress on Advanced Applied Informatics, IIAI-AAI 2023
Pages
287-292
ISBN
9798350324228
Date Issued
2023-01-01
Author(s)
Abstract
This research paper proposes a novel machine learning (ML) system to automate the generation of myocardial infarction (MI) classifiers using lead I electrocardiogram (ECG) signals. The proposed system comprises three primary processes: data preprocessing, automated model generation, and model evaluation. The automated model generation process executes two procedures, namely, model generation and model optimization. In the model generation procedure, the system utilizes ten feature engineering algorithms and nine machine learning models to generate 90 models automatically. The generated models are then fine-tuned by the model optimization procedure through meta learning and hyperparameter optimization algorithms. The proposed system's significant advantage is that users only need to provide the dataset and prediction target of an application, and the embedded machine learning algorithms automatically generate an optimal model. The system is easy to use and an end-to-end modeling solution that can be applied to various fields without requiring machine learning expertise. To validate the system's effectiveness, the Physikalisch-Technische Bundesanstalt (PTB) open electrocardiogram (ECG) dataset is utilized to generate a myocardial infarction (MI) location classifier using the lead I ECG signal of the dataset. Among the 90 models generated, the random forest model with the system-identified parameters achieved the best performance, achieving an accuracy, sensitivity, specificity, F1 score, and classification accuracy of 99.90%, 99.46%, 99.94%, 99.55%, and 99.40%, respectively. The results demonstrate that the proposed automated ML system is an efficient and effective approach for non-ML experts to leverage AI technology to solve their applications in an easy-to-use, end-to-end manner.
Subjects
Automated Machine Learning | Classification | Hyperparameter Learning | Model Generation | Myocardial Infarction | Random Forest
Publisher
Institute of Electrical and Electronics Engineers Inc.
Type
conference paper