Traditional Programming is running data and programs on a computer to produce output and Machine Learning is running data and output on a computer to create a program. The program can be used in traditional programming. Machine learning is a science that allows computers to act without explicit programming[25]. In the past decade, machine learning has brought us self-driving cars, useful speech recognition, efficient web search, and a vast understanding of the human genome. Machine learning is so ubiquitous today that you might be using it dozens of times a day without even knowing it. In machine learning, there are a large number of linear classifiers, nonlinear classifiers, statistics-based classifiers, and probabilistic classifiers, etc. These classification algorithms usually have the characteristics of fast training speed and clear action mechanism[37]. This section mainly introduces the Decision Tree, Gradient Boosting Decision Tree (GBDT), Random Forest, Logistic regression, SVM, and K-Nearest Neighbor (KNN) algorithm.

（1）Decision tree

The decision tree is one of the simplest and most useful machines learning structures. Decision trees are an approach to machine learning. Decision tree generation algorithms include ID3, C4.5, and C5.0. A decision tree is a tree-shaped structure in which each internal node represents a judgment on an attribute, each branch represents the output of a judgment result, and finally, each leaf node represents a classification result[38]. Decision tree learning methods are commonly used in data monitoring. The model is targeted at predicting the quality of a target available based on input available. Use decision trees to classify data and build models that describe important class variables for machine learning and pattern recognition. The output of the classification problem is treated as a pattern of all observed values of the terminal node[39]. The method of establishing classification models is mainly divided into two steps. First, the classification model is established based on training data. Second, the accurateness of the model is checked and classified by the new data model. The category labels shown here take the form of discrete values, such as "Yes" or "No", "Dangerous" or "Safe", etc.

（2）Gradient lifting tree

Gradient Boosting is a method of boosting. Its main idea is that every time a model is established, the gradient descent direction of the model loss function is established before. Gradient boosting can present a more efficient algorithm and a training dataset faster. It can benefit from regularization methods and advises different parts of the algorithm and generally improves the effectiveness of the algorithm by reducing overfitting. The function of the loss is that the performance of the model is believed that the lower the amount of damage, the better the efficiency[40]. The loss function used in the gradient boosting algorithm may vary depending on the type of problem being solved, but many can support the standard loss function and define its own functions. For example, use the log decay function that can use the regression square loss difference function and classification.

The GBDT algorithm uses the decision tree as weak learning of the gradient boosting model[6][41]. First, each weak student will weigh. The program will run a specific set of training images. Then, given the results, it will re-weigh the array of weaker students. If a much better estimate is made than before, the effect will be raised accordingly. GBDT is a greedy algorithm that can quickly fit training data sets. This allows the output of subsequent models to be added and residuals in the prediction to be "corrected". For example, the classification of binary images. The decision tree is constructed in a greedy way and the best split point is selected according to the Gini index. GBDT algorithm can use the "ensemble" method of gradient boosting in almost any type of ML project, from HSI recognition to the classification of user recommendations or natural language analysis.

（3）Random forests

A random forest is a decision tree that is generated using a set of the large tree and mathematical algorithms that give equal probabilities of trees in a given distribution. The random tree is most commonly generated with a random tree generator. Each tree produces category predictions, and the category with the most votes become the prediction of the model[42]. Two methods, such as bagging and feature random selection are used randomly to ensure that each individual tree in the random forest does not have too much to do with the behavior of another tree.

Random forest algorithm creates decision trees on data samples and then gets the prediction from each of them and finally selects the best solution employing voting. The performance of decision trees is easily unstable due to the quality of samples. Random forest allows each individual tree to be replaced by random sampling from the data set. Similarly, to obtain different decision trees in this way, this process is called bagging. They reduce errors using greedy algorithms that want to split variables[42]. For example, decision trees can have many structural similarities with bagging and instead have a high correlation in their predictions. It should be noted that the use of bagging does not divide the training data into multiple parts, nor does it train each tree into different data blocks. In the models in this series, the difference between the decision trees is even greater, and ultimately the higher the correlation between the decision trees. As the Bagged decision trees are constructed and can calculate how much the error function drops for a variable at each split point. Therefore, the final decision tree in random forests not only needs to be trained for different data sets later, but it also needs to use different functions for decision making.

（4）Logistic regression

In statistics, logical models are usually used to model the probability of a specific category or event. Logistic regression is a machine learning method for solving binary (0 or 1) problems, which is used to estimate the probability of something[39]. Logical regression assumes that the dependent variable follows a Bernoulli distribution. Logistic regression easily introduces nonlinear components through the sigmoid function, so it can easily deal with 0/1 classification problems. Logical regression assumes that the dependent variable follows a Bernoulli distribution, while linear regression assumes that the dependent variable follows a Gaussian distribution. First of all, need to introduce the Sigmoid function, also known as the Logistic function. Similar to linear regression, which assumes that the data follows the linear function, logistic regression also uses the S-shaped function to model the data, as follows:

Logical regression can be used as a classifier only when a decision threshold is introduced. The setting of the threshold is an important element of logistic regression and depends on the classification problem itself.

（5）Support vector machine

Support vector machine is a supervised machine learning algorithm. It is a two-class classification model. Its basic model is defined as the linear classifier with the largest interval in the feature space. Its learning strategy is to maximize the interval, which can finally be transformed into the solution of a convex quadratic programming problem. The basic idea is to find the farthest boundary in each category and determine whether the sample belongs to a region or other region by solving the basic mathematical function given the coordinates of each sample. In addition to linear classification, support vector machines can use so-called nuclear techniques, and their inputs are implicitly mapped into high-dimensional feature spaces for effective nonlinear classification. The SVM algorithm is implemented using the kernel, and the hyperplane learning in linear SVM is accomplished by using some linear alleged conversion problems[43].

Among the traditional machine learning algorithms, SVM has some irreplaceable advantages. For example, SVM learns faster and performs better when the number of samples is limited, making the algorithm well suited for text classification problems, so for such cases, up to thousands of data sets with labeled samples can usually be accessed

（6）K - nearest neighbor

The k-nearest neighbors (KNN) algorithm is a simple, easy-to-implement supervised machine learning algorithm that can be used to solve both classification and regression problems. The KNN algorithm belongs to the overall scope of supervised learning, and its classification principle is based on the simple thing that is accepted as true or as certain to happen, without proof that similar things are always closer. A method in K-mean clustering vector quantization and the K-means algorithm is used to cluster to determine and can be used for Regression as well as for classification but mostly it is used for the classification problems, it is a non-parametric algorithm, which means it does not make any assumption on underlying data [44]. The K-NN algorithm estimates seminaries between new case and data and available case and keeps the new case and data in categories that are similar to the available categories. K-NN algorithm stores all the available data and classifies a new data point based on the similarity, in other words, KNN algorithm at the training phase just stores the dataset and when it gets new data, then it classifies that data into a category that is much similar to the new data.

The KNN algorithm has the characteristics of being very easy to explain and understand, and despite its simplicity, it can still produce highly competitive results. At the same time, KNN algorithm is a non-parametric algorithm, which does not need to make any assumptions about data distribution, which makes KNN more advantageous in processing special data. Higher-dimensional data can cause the accuracy of the KNN algorithm to decline because the nearest element is almost indistinguishable from the farthest element.