## Table of Contents

**PCA in Machine Learning**

Principal Component Analysis is a type of unsupervised learning algorithm that is utilized in the field of machine learning for the purpose of dimensionality reduction. Using orthogonal transformation, this statistical method transforms the observations of interrelated characteristics into a series of linearly uncorrelated data. This is done by transforming the correlated features into orthogonal coordinates. These newly remodeled characteristics have been given the name Principal Components. It is one of the most common tools utilized in the process of exploratory data analysis and predictive modeling. It is a method for extracting robust patterns from a given dataset by taking steps to minimize the amount of variation in the data. In most cases, the principal component analysis will look for a lower-dimensional surface to project higher-dimensional data. Learning Machine Learning

The principal component analysis (PCA) works by taking into account the variance of each characteristic. This is done because a high attribute reveals a good split between the classes, which minimizes the dimensionality. Image processing, movie recommendation systems, and optimizing power distribution across a variety of communication channels are some real-world applications of principal component analysis (PCA). Because it is a method for extracting features, it takes into account the most significant factors while disregarding the less significant ones.

**Ideas Behind PCA in ****Machine Learning**

The PCA algorithm is founded on a number of mathematical ideas, including the following:

- The concepts of Variance and Covariance
- Eigenvalues and Eigen factors

**Terminology in Principal Component Analysis**

**The following is a list of some popular terminology used in the PCA algorithm:**

**Dimensionality**

The term “dimensionality” refers to the number of distinct features or variables that are included in a certain dataset. The number of columns that are included in the dataset is a much simpler indicator of this.

**Correlation**

The term “correlation” refers to the degree to which two different variables are connected to one another. For example, if one variable is updated, it will cause the other variable to also change. The correlation value can be anywhere from minus one to plus one. In this case, we get a value of -1 if the variables in question are inversely proportional to one another, and we get a value of +1 if the variables in question are directly proportional to one another.

**Orthogonal**

It establishes that the variables in question are not related to one another in any way, leading to the inevitable conclusion that there is no correlation between the two sets of data.

**Eigenvectors**

Eigenvectors are generated whenever there is a square matrix M and a vector that is not zero is provided. If a scalar multiple of v, Av, exists, then v will be an eigenvector in that case.

**Covariance Matrix**

The term “covariance matrix” refers to a matrix that contains the covariance that exists between a pair of variables.

**PCA’s Principal Components**

As was just said, the Principal Components are either the newly modified features or the results of the principal component analysis. The number of these PCs is either less than the total number of the original characteristics that were included in the dataset or it is the same. The following is a list of some of the characteristics of these primary components:

- The linear combination of the initial features needs to be the primary component.
- These components are orthogonal, which means that there is no correlation between any two of the variables that are being considered.
- Since the relevance of each component lessens as the number of components increases from one to n, this indicates that component number one has the highest level of importance, while component number n will have the lowest level of importance.

**Phases of the PCA algorithm**

**Acquisition of the dataset**

First things first, we need to take the dataset that was provided to us and split it into two parts: X and Y. X will serve as our training set, while Y will serve as our validation set.

**Fixing the Data into a Structure**

The next step is to create a structure that represents our dataset. In this manner, the two-dimensional matrix of the independent variable X will be represented by us. In this table, each row represents a different data item, and each column represents a different feature. The dimensions of the dataset are indicated by the number of columns.

**Creating a standard for the data**

The standardization of our dataset will take place in this step. For example, in a specific column, the elements that have a higher variation are considered to be more essential than the features that have a smaller variance.

If the significance of features is not reliant on the degree to which those features vary, then we shall divide each data item in a column by the column’s standard deviation. At this point, we will refer to the matrix as Z.

**The Covariance of Z Is Calculated**

In order to compute the covariance of Z, we will begin by transposing the matrix Z that we have just created. Following the transposition, we shall carry out the multiplication by Z. The covariance matrix of Z is going to be the matrix that is output.

**Determine the EigenValues and EigenVectors**

Now that we have the resultant covariance matrix Z, we need to determine its eigenvalues and eigenvectors. The directions of the axes that contain the most information are referred to as eigenvectors or covariance matrices. Additionally, the eigenvalues are referred to as the coefficients of these eigenvectors.

**The EigenVectors Are Being Sorted**

During this stage of the process, we will collect all of the eigenvalues and arrange them in descending order, which means from the greatest to the least significant. And at the same time, arrange the eigenvectors in the appropriate order in the matrix P of eigenvalues. P* will be the name given to the final matrix that was created.

**Figuring out the latest features, also known as Principal Components**

In this section, we will compute the newly added features. In order to accomplish this, we are going to multiply the P* matrix by the Z. Each observation in the resultant matrix Z* is the linear combination of the features that were present in the original data. Independent of one another, the columns of the Z* matrices can be arranged in any order.

**Eliminate any features from the new dataset that are less significant or irrelevant**

The new feature set has been implemented, therefore from this point on, we will select what to keep and what to get rid of. It means that in the new dataset, we will only maintain the features that are relevant or important, and we will eliminate the features that are not relevant or important.

**PCA’s Potential Uses**

We can now turn our attention to the applications of PCA now that we have finished gaining knowledge of what PCA is and why it is effective. PCA’s most common use is in reducing the number of dimensions being considered. If you have data with a high number of dimensions, you can use principal component analysis (PCA) to reduce the dimensionality of those data such that the vast majority of the variance that exists in your data across those many dimensions is represented by a smaller number of variables.

In all types of analysis, from quantitative finance to neurology, principal component analysis (PCA) is employed extensively. PCA is utilized in a wide variety of different fields and industries. The following is a list of some of the most important uses of PCA:

**NeuroScience**

In the field of neuroscience, a method known as spike-triggered covariance analysis is a form of Principal Components Analysis that is used to determine the specific qualities of a stimulus that raise the likelihood of a neuron producing an action potential.

In addition to this, PCA can be utilized to determine the identification of a neuron based on the form of its action potential.

PCA is a technique for reducing the number of dimensions that is utilized to detect coordinated actions of huge neural ensembles. During phase transitions in the brain, it has been utilized in the process of determining collective variables, also known as order parameters.

**Quantitative Finance**

PCA is a method that can be used to reduce the number of dimensions in a difficult task. Let’s say that a fund manager’s portfolio consists of 200 different stocks. In order to do a quantitative analysis of these stocks, a stock manager will need a co-relational matrix with a dimension of 200 squares by 200 squares, which makes the problem very difficult to solve.

However, if he were to extract 10 Principal Components that best represent the variance in the stocks, this would lessen the complexity of the problem while still describing the movement of all 200 stocks. These 10 components would be chosen based on their ability to best represent the variance. The following are some further applications of PCA:

- Performing an examination of the form of the yield curve
- Portfolios that are hedged with fixed income
- The application of models pertaining to interest rates
- Speculating the returns of a portfolio
- The development of algorithms for asset allocation
- Developing algorithmic strategies for long-short equities trading

**Image Compression**

Image compression is another application for PCA.

**Facial Recognition**

- One of the most important methods in computer vision is known as EigenFaces Face Recognition, which may be used to identify people’s faces.
- Sirovich and Kirby (1987) demonstrated that principal component analysis (PCA) may be applied to a group of digitized facial photos in order to derive a list of fundamental facial characteristics.
- Because PCA is used to construct the set of EigenFaces, it is considered to be an integral part of the EigenFaces methodology.
- The Eigenface technique lessens the statistical complexity of representing faces in images.
- The accuracy of face recognition has been improved by other studies by utilizing Wavelet, PCA, and Neural Networks in conjunction with one another.

**Additional Applications**

**PCA has additionally been put to use in a wide variety of other applications, some of which are described below :**

- The principal component analysis (PCA) was applied to medical data in order to demonstrate an association between cholesterol and low-density lipoprotein.
- The principal component analysis (PCA) technique has been applied to HVSR (horizontal to vertical spectral ratio) data for the purpose of seismically characterizing regions that are prone to earthquakes.
- PCA has been utilized in the process of detecting and visually representing attacks on computer networks.
- Anomaly Detection is one application that has made use of PCA.