Brief Review — Analysis of Training Optimization Algorithms in the NARX Neural Network for Classification of Heart Sound Signals
NARX
3 min readMar 16, 2024
Analysis of Training Optimization Algorithms in the NARX Neural Network for Classification of Heart Sound Signals
NARX, by High institute of Engineering and Technology, and Aswan University
2022 IJSER (Sik-Ho Tsang @ Medium)Heart Sound Classification
2013 … 2023 [2LSTM+3FC, 3CONV+2FC] [NRC-Net] [Log-MelSpectrum + Modified VGGNet] [CNN+BiGRU] [CWT+MFCC+DWT+CNN+MLP] [LSTM U-Net (LU-Net)]
==== My Other Paper Readings Are Also Over Here ====
- Nonlinear autoregressive networks with exogenous inputs (NARX) network is proposed for the classification of many different features extracted from labelled PCG signals.
Outline
- NARX
- Results
1. NARX
1.1. Overall Framework
- The Overall Framework is shown above: Feature extraction > NARX > Classification.
1.2. Feature Extraction
- A total of 27 features are extracted and categorized as shown in Table I.
- (Please kindly read the paper for more details of eahch feature.)
1.3. NARX Neural Network
- The output y(t) is also used as input for training the neural network similar to the RNN concept.
- This open-loop network is used during the training process because of the availability of the true past values of the time series.
- MSE is used as loss function.
- 3 types of optimization algorithms are tried: Scaled Conjugate Gradient (SCG), Levenberg Marquardt (LM), Bayesian Regularization (BR).
- The neural NARX networks are converted to closed-loop network, which is beneficial for multi-step-ahead prediction after the training process [28,29].
2. Results
- PhysioNet dataset is used.
In general, the NARX classifier provides comparable performance when trained with the BR and LM algorithms. However, it gives low performance values when trained with the SCG algorithm.
