**2.1 Machine learning (ML) processes**

Machine learning (ML) builds the data investigative algorithms to extort characteristics from the data. Inputs to machine learning (ML) algorithms consist of patient 'characters' and occasionally therapeutic effects of concern. A patient's characters generally contain bottom line data, for example, gender, age, disease history, and also disease explicit data, for instance, gene expressions, analytical imaging, electrophysiological data (EP) test, objective test results, medication, and medical symptoms. In addition to the attributes of the patients medical results are frequently composed for medical investigation. These contain syndrome pointers, patients' endurance periods, and quantitative syndrome stages such as the size of tumor. Here *j* th characteristic of the *i* th numbers of patient is denoted by *Pij* and *Qi* is representing the effect of concern. Regarding whether to integrate the results, machine learning (ML) algorithms can also be separated into two main types: supervised learning and unsupervised learning. One more type is also available

#### **Figure 1.**

*The road plan from generation of medical data, during natural language processing (NLP) data improvement and machine learning (ML) data investigation.*

named as semisupervised learning. **Figure 2** represents all these three types of learning procedures. Unsupervised learning is also identified for feature removal, whereas supervised learning is appropriate for analytical representation by constructing several interactions involving patient individuality (input) and result of concern (output). In recent times, semisupervised learning has been projected as a hybrid involving supervised learning and unsupervised learning, which is appropriate for circumstances wherever the effect is omitted for definite issues.

There are two major unsupervised learning techniques available such as (i) principal component analysis (PCA) technique and (ii) clustering technique. Principal component analysis is basically for element reduction, mainly while the characteristic is documented in a huge number of elements, such as the number of genes in a genome-mixt connection revise. Principal component analyses (PCA) project the data on a small number of principal component (PC) guidelines, without trailing in excess of information regarding the issues. Occasionally, PCA is used to decrease the element of the data, after which clustering technique is used to fraction the issues. All these fraction issues with related characteristics are gathered together, without applying any result information. This algorithm's result output helps the cluster tags for the patients throughout maximizing as well as minimizing the parallel of the patients and also involving the clusters. These accepted clustering algorithms contain (i) Gaussian mixture clustering, (ii) K-means clustering, and (iii) hierarchical clustering. Alternatively, supervised learning reflects on the topics' outcomes in cooperation with their characteristics and goes via a definite training procedure to find out the finest outputs connected through the inputs, which are nearby the standard outcomes. Generally, the formulations of output contrast through the concern outcomes. Such that, the outcome can be the possibility of receiving an exact clinical result, the projected value of a disease stage or the projected endurance time. Evaluated by unsupervised learning and supervised learning, which offers extra clinically applicable results; therefore Artificial Intelligence (AI) relevance in healthcare system most regularly apply supervised learning. Unsupervised learning may be applied as a component of the preprocessing stage to or find out subgroups or decrease dimensionality, which consecutively makes summarizing supervised learning stage more capable. Appropriate methods contain logistic regression, linear regression, decision tree, naïve Bayes, random forest, discriminate analysis, nearest neighbor, neural network, and support vector machine (SVM). Neural network and SVM are the most accepted supervised learning methods in healthcare applications [5]. The mechanisms of neural networks and support vector machine (SVM) techniques process together with relevant examples in the cardiovascular disease, neurological disease, and cancer.

**Figure 2.** *Representation of (A) unsupervised learning, (B) supervised learning, and (C) semisupervised learning.*

**125**

the *i*

*i*

*Application of Artificial Intelligence in Modern Healthcare System*

Neural network is basically known as the expansion of linear regression for confining the difficult nonlinear relationships dividing the input parameters and outcome data. In this neural network, the relations involving the input parameters and the outcome are represented throughout the multiple unknown layer grouping of preindividual functional. The aim is to calculate approximately the weights via input data and also the outcome data so that the average error involving the outcome and their calculation is reduced. Here, this technique is described via following some examples. Neural network was used in stroke diagnosis [6], where the input parameters were given as *Xi*1, …, *Xip* and *p* = 16 stroke-related symptoms, together with acute confusion, problem of vision and mobility, paresthesia of the leg or arm, etc. *Yi* represents the binary outcome, where *Yi* = 1/0 represents that

th patient has or does not have stroke. The output factor of importance is the

In this equation, *w*10 and *w*20 are not equal to zero, where *Xij*, *fk* = 0; *fk*s and h are prespecified functions, which indicate that the weighted grouping influences the

The instruction's aim is to find out the weight of *wij,* which can minimize the

standard optimization algorithms, for instance, local quadratic estimate or gradient decline optimization, which are integrated in both R and MATLAB software. The latest data were issued from the similar population and the results of *wij* are also applied to calculate the outcomes rooted in their particular characters [7]. This is the same as methods have been applied to identify cancer treatment [8], where the input efforts and outcomes are the principal components (PC) predictable from 6567 genes and the tumor groups. A neural network was applied to identify breast cancer, where the inputs represent the surface information from mammographic images and where the outcomes are tumor indicators [9]. Another problematical neural network model was analyzed to identify Parkinson's disease derived where the input parameters are motor and nonmotor indications and neuroimages [10].

The supporting vector machine is mostly applied for categorizing the topics into two different clusters, where the result *Yi, Yi* = −1 or 1 indicates whether the

th patient is in set 1 or 2 correspondingly. This procedure can be completed for circumstances with more than 2 sets. The fundamental hypothesis is that the subject matters can be divided into two different groups via a decision boundary distinct on

> *ai* = Σ *j*=1 *p*

comparative implication on moving the outcome between the others. If *ai* > 0,

th patient is categorized to group 1, that is, *Yi* = −1; and if *ai* < 0, the patient is categorized to group 2, that is, *Yi* = 1. Furthermore, assuming that the new patients come from the same population, the resulting *Wj* can be applied to classify these new patients based on their traits. An important property of SVM is that the determination of the model parameters is a convex optimization problem so the solution

*<sup>n</sup>* (*Yi* − *ai*)<sup>2</sup>

*w*1*l Xil* + *w*10) + *w*20} (1)

. The minimization can be done via

*wj Xij* + *b* (2)

th characteristic to mark edits'

possibilities of stroke (*ai),* which represents the equation given below:

*w*2*l fk*(Σ *l*=1 *p*

disease threat as a whole. **Figure 3** represents the neural network system.

*ai* = *h*{Σ K=1 *D*

calculation in accuracy given by Σ*i*=1

**2.3 The support vector machine (SVM)**

the characteristics *Xij*, which can be represented as:

where *wj* represents the weight put on the *j*

*DOI: http://dx.doi.org/10.5772/intechopen.90454*

**2.2 Neural network**

the *i*

## **2.2 Neural network**

*Alginates - Recent Uses of This Natural Polymer*

named as semisupervised learning. **Figure 2** represents all these three types of learning procedures. Unsupervised learning is also identified for feature removal, whereas supervised learning is appropriate for analytical representation by constructing several interactions involving patient individuality (input) and result of concern (output). In recent times, semisupervised learning has been projected as a hybrid involving supervised learning and unsupervised learning, which is appro-

priate for circumstances wherever the effect is omitted for definite issues. There are two major unsupervised learning techniques available such as (i) principal component analysis (PCA) technique and (ii) clustering technique. Principal component analysis is basically for element reduction, mainly while the characteristic is documented in a huge number of elements, such as the number of genes in a genome-mixt connection revise. Principal component analyses (PCA) project the data on a small number of principal component (PC) guidelines, without trailing in excess of information regarding the issues. Occasionally, PCA is used to decrease the element of the data, after which clustering technique is used to fraction the issues. All these fraction issues with related characteristics are gathered together, without applying any result information. This algorithm's result output helps the cluster tags for the patients throughout maximizing as well as minimizing the parallel of the patients and also involving the clusters. These accepted clustering algorithms contain (i) Gaussian mixture clustering, (ii) K-means clustering, and (iii) hierarchical clustering. Alternatively, supervised learning reflects on the topics' outcomes in cooperation with their characteristics and goes via a definite training procedure to find out the finest outputs connected through the inputs, which are nearby the standard outcomes. Generally, the formulations of output contrast through the concern outcomes. Such that, the outcome can be the possibility of receiving an exact clinical result, the projected value of a disease stage or the projected endurance time. Evaluated by unsupervised learning and supervised learning, which offers extra clinically applicable results; therefore Artificial Intelligence (AI) relevance in healthcare system most regularly apply supervised learning. Unsupervised learning may be applied as a component of the preprocessing stage to or find out subgroups or decrease dimensionality, which consecutively makes summarizing supervised learning stage more capable. Appropriate methods contain logistic regression, linear regression, decision tree, naïve Bayes, random forest, discriminate analysis, nearest neighbor, neural network, and support vector machine (SVM). Neural network and SVM are the most accepted supervised learning methods in healthcare applications [5]. The mechanisms of neural networks and support vector machine (SVM) techniques process together with relevant examples

in the cardiovascular disease, neurological disease, and cancer.

*Representation of (A) unsupervised learning, (B) supervised learning, and (C) semisupervised learning.*

**124**

**Figure 2.**

Neural network is basically known as the expansion of linear regression for confining the difficult nonlinear relationships dividing the input parameters and outcome data. In this neural network, the relations involving the input parameters and the outcome are represented throughout the multiple unknown layer grouping of preindividual functional. The aim is to calculate approximately the weights via input data and also the outcome data so that the average error involving the outcome and their calculation is reduced. Here, this technique is described via following some examples. Neural network was used in stroke diagnosis [6], where the input parameters were given as *Xi*1, …, *Xip* and *p* = 16 stroke-related symptoms, together with acute confusion, problem of vision and mobility, paresthesia of the leg or arm, etc. *Yi* represents the binary outcome, where *Yi* = 1/0 represents that the *i* th patient has or does not have stroke. The output factor of importance is the possibilities of stroke (*ai),* which represents the equation given below:

$$\mathfrak{a}\_{i} = h \left\{ \mathop{\sum}\_{\mathbf{K}=1}^{D} \boldsymbol{w} \mathfrak{U} \boldsymbol{f} \mathbb{k} \left( \mathop{\sum}\_{l=1}^{p} \boldsymbol{w} \mathfrak{U} \boldsymbol{1} \boldsymbol{X}\_{il} \star \boldsymbol{w}\_{10} \right) \star \boldsymbol{w}\_{20} \right\} \tag{1}$$

In this equation, *w*10 and *w*20 are not equal to zero, where *Xij*, *fk* = 0; *fk*s and h are prespecified functions, which indicate that the weighted grouping influences the disease threat as a whole. **Figure 3** represents the neural network system.

The instruction's aim is to find out the weight of *wij,* which can minimize the calculation in accuracy given by Σ*i*=1 *<sup>n</sup>* (*Yi* − *ai*)<sup>2</sup> . The minimization can be done via standard optimization algorithms, for instance, local quadratic estimate or gradient decline optimization, which are integrated in both R and MATLAB software. The latest data were issued from the similar population and the results of *wij* are also applied to calculate the outcomes rooted in their particular characters [7]. This is the same as methods have been applied to identify cancer treatment [8], where the input efforts and outcomes are the principal components (PC) predictable from 6567 genes and the tumor groups. A neural network was applied to identify breast cancer, where the inputs represent the surface information from mammographic images and where the outcomes are tumor indicators [9]. Another problematical neural network model was analyzed to identify Parkinson's disease derived where the input parameters are motor and nonmotor indications and neuroimages [10].

### **2.3 The support vector machine (SVM)**

The supporting vector machine is mostly applied for categorizing the topics into two different clusters, where the result *Yi, Yi* = −1 or 1 indicates whether the *i* th patient is in set 1 or 2 correspondingly. This procedure can be completed for circumstances with more than 2 sets. The fundamental hypothesis is that the subject matters can be divided into two different groups via a decision boundary distinct on the characteristics *Xij*, which can be represented as:

$$\mathcal{a}\_i = \sum\_{j=1}^p \omega \nu\_j X\_{ij} + b \tag{2}$$

where *wj* represents the weight put on the *j* th characteristic to mark edits' comparative implication on moving the outcome between the others. If *ai* > 0, the *i* th patient is categorized to group 1, that is, *Yi* = −1; and if *ai* < 0, the patient is categorized to group 2, that is, *Yi* = 1. Furthermore, assuming that the new patients come from the same population, the resulting *Wj* can be applied to classify these new patients based on their traits. An important property of SVM is that the determination of the model parameters is a convex optimization problem so the solution

is always global optimum. Additionally, many obtainable rounded optimization technique applications are readily available for the SVM performance. SVM has been widely applied in healthcare research. For example, SVM was used to recognize imaging biomarkers of psychiatric and neurological disease [11]. SVM was also applied in cancer diagnosis [12]. SVM and other statistical methods can also be used to reach early detection of Alzheimer's syndrome [13]. SVM was applied to analyze the power of an offline human and device interface, which can control the upperlimb prostheses [14].
