Table 5.

Analysis of some well-known feature extraction methods of HARS.

Method	Idea	Advantage	Disadvantage
PCA	It is a linear method and consists of converting the main features (generally interdependent) into new features that are not interdependent and depend on the data's scale.	(i) Returns the main features to a low-dimensional space. (ii) Elimination of the central parts leads to lower variance and increased accuracy.	(i) The principal components are not always easy to interpret. (ii) Changes within the class.
LDA	Features extracted through linear conversion to find the variables' linear composition, which is the best representation of the data.	(i) Minimizes changes within the class relative to principal component analysis. (ii) Converts the main features to a new space with lower dimensions. (iii) Maximizes segregation between classes.	(i) Relies on a complex model containing the correct number of components. (ii) Limits flexibility when using complex datasets. (iii) Lack of covariance matrix within the same class. (iv) The possibility of insufficient data to estimate the conversions in the separation of classes.
ICA	This method finds independent components such as main features expressed as a linear combination of components.	(i) Solution to solve the problem of blind source separation. (ii) Effective for describing local features.	(i) Suitable for non-Gaussian data. (ii) Computationally expensive. (iii) Unsuitable for online algorithms.
FA	The main features can be grouped according to their correlation.	(i) The features of each group are strongly correlated. (ii) Quantitative communication between the features of different groups.	(i) Investigates the factors and finds the most effective ones.
DLF	The salient features of the raw sensor data can be extracted automatically, without relying on handcrafted features.	(i) Ability to automatically learn from unauthorized and, in some cases, unlabeled raw sensor data. (ii) These methods offer different capabilities for processing sensor current.	(i) Searches for optimal solutions. (ii) High calculation time due to setting the above parameters.