Brief Review — An Optimal Approach for Heart Sound Classification Using Grid Search in Hyperparameter Optimization of Machine Learning
MFCC + k-NN / RF / ANN / SVM + Grid Search
An Optimal Approach for Heart Sound Classification Using Grid Search in Hyperparameter Optimization of Machine Learning
MFCC + k-NN / RF / ANN / SVM + Grid Search, by Kumoh National Institute of Technology, Telkom University, Meta Heart Co., Ltd.
2023 MDPI J. Bioengineering, Over 20 Citations (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)] [DL Overview]
==== My Other Paper Readings Are Also Over Here ====
- MFCC features are extracted and used as input of classification methods: k-NN, RF, ANN, SVM.
- Grid search is performed to fine-tune the hyperparameters of k-NN, RF, ANN, SVM.
Outline
- Grid Search to Fine-Tune the Hyperparameters of k-NN, RF, ANN, SVM
- Results
1. Grid Search to Fine-Tune the Hyperparameters of k-NN, RF, ANN, SVM
1.1. MFCC Feature Extraction
- The number of MFCC features used in this study is 13, 25, 42.
1.2. k-NN
The best parameter, as well as the best k value selection for varied values of k (k = 1, 3, 5, 7, . . . , 31), the distance matrix, including Euclidean, Minkowski, and Chebyshev, were selected throughout the optimization process using the grid search method.
1.3. SVM
The selection of γ parameters (from 0.1 to 0.001) and the regularization of C parameter selection for SVM (from 1 to 1×10³) was evaluated.
1.4. ANN
- The model consists of 2 hidden layers with 100 nodes for each hidden layer. ReLU is used except Sigmoid is used at output.
The optimization of ANN was performed using a grid search to select the best optimizer among Adam, Nadam, SGD, and RMSprop optimizers and the best learning rate (from 0.1 to 0.001).
1.5. RF
- The grid search method determines the optimum number of trees (50, 100, 150, and 200) and the best criterion (gini and entropy) that yields the best performance result as the RF’s optimal parameter
2. Results
2.1. k-NN, RF, ANN, SVM Classification Performance
- 42 MFCC features are the best number of features.
For both datasets, the Minkowski distance with an optimal value of k = 1 was chosen as the K-NN algorithm’s best parameter since it obtained the optimal classification performance.
2.2. SOTA Comparisons
The proposed K-NN model with 42 MFCC features outperformed the previous studies.
