Shapelet Feature Learning Method of BCG Signal Based on ESOINN

Abstract

Ballistocardiogram (BCG) signal is an effective information that can be used to diagnose cardiovascular disease. This paper analyzes a method of learning the Shapelet feature of BCG signal based on ESOINN. Firstly, the original BCG signal is pre-learned using an enhanced self-organizing incremental unsupervised neural network (ESOINN); Then, it's transformed by the shapelet transform algorithm; Finally, the feature selection method is used to select the shapelet feature from the candidate set, and carry out the training of the classifier. The results show that the method can learn the better quality shapelet candidate set, and greatly reduce the number of candidate sets. In addition, the learning time complexity of shapelet features is greatly reduced, and the accuracy of the model is improved.