Some understandings on machine learning algorithm (decision tree, SVM,KNN nearest neighbor, Random forest, naive Bayesian, logistic regression)

I. Decision TREE

Set an initial particle, starting at that point and branching out. (because the initial particle may fall on the boundary value, there may be problems with fitting at this point.)

Second, Svm

SVM is the best classification algorithm in addition to deep learning before the advent of deep learning. It has the following characteristics:

(1) It can be applied to the linear (regression problem) classification, but also to the nonlinear classification;

(2) By adjusting the setting of kernel function parameters, the data set can be mapped to multidimensional plane, fine-grained, so that its characteristics from two-dimensional to multi-dimensional, will be on the two-dimensional linear irreducible problem into a multi-dimensional linear can be divided, Finally, we look for an optimal cutting plane (which is equivalent to finding an optimal solution based on the decision number), so the classification effect of SVM is better than that of most machine learning classification methods.

(3) By setting other parameters, SVM can also prevent the problem of overfitting.

iii. Random Forest

　　In order to prevent overfitting, a random forest is equivalent to several decision trees.

Four, KNN nearest neighbor

Since KNN has to traverse all the remaining points each time it looks for the next closest point to it, the algorithm is expensive.

V. Naive Bayes

To push the probability that the occurrence of event a occurs under B (where events A and B can be decomposed into multiple events), you can calculate the probability of event a occurring under the probability of event B, and then compute the result by Bayes theorem.

Vi. Logistic regression

(discrete variable, two- classification problem, only two values 0 and 1)

Some understandings on machine learning algorithm (decision tree, SVM,KNN nearest neighbor, Random forest, naive Bayesian, logistic regression)