1. A Comparative Study on Classification of Sleep
Stage Based
on EEG Signals Using Feature Selection and Classification
Algorithms 4 different classes of features (temporal, nonlinear, frequency-based and entropy). A total of 21 feature algorithms were used; 10 in the time category, 5 in the nonlinear category, 2 in the frequency category, and 4 in the entropy category. 41 eigenvalues were obtained from 21 eigenvalues. Time domain: zero-crossing times; Hjorth coefficient; statistical features include: nonlinear features 5: Petrosian fractal dimension (PFD) Mean teager energy (MTE) MTE Mean energy (ME) Mean curve length (MCL) MCL Hurst exponent (H) Time-frequency domain features: Wigner-Ville distribution (WV): WV-1-WV-4 Discrete wavelet transform (DWT): D3-1-D5-4 Entropy features: REn (Rayleigh entropy); PEn arrangement entropy; SpEn (spectrum Entropy); ApEn (Approximate Entropy);
Eigenvalue
Feature selection algorithm: 1.Fast correlation based filter (FCBF)
2.mRMR algorithm()
3.T-test algorithm (T-test algorithm)
4.ReliefF algorithm
5.Fisher score algorithms (Fisher score)
6.Minimum Redundancy maximum relevance (MRMr); minimum redundancy maximum relevance
(mRMR);
Feature classification algorithm : 1. Decision trees (DT)
2. Feed-forward neural network (FFNN) (feedforward neural network)
3. Radial basis network (RBF) radial basis network
4. Support vector machine (SVM)
5 .Random forest algorithm (RF) (random forest)
performance evaluation method : 1.Classification accuracy, sensitivity and specificity (classification accuracy, sensitivity and
specificity evaluation)
2.k-Fold cross-validation (k compound cross-validation)
conclusion : Based on the results, the RF algorithm is considered to be the best algorithm for success. Subsequently, radio frequency, SVM, discrete cosine transform and radial basis function algorithms were found to be effective in different feature clusters.
The DT algorithm is the algorithm with the least computation time. Generally speaking, the algorithm is followed by radio frequency algorithm, SVM algorithm, fast Fourier transform neural network algorithm and radial basis function algorithm.