Shulou(Shulou.com)06/01 Report--

What this article shares with you is about how to solve big data's multicollinearity problem. The editor thinks it is very practical, so I share it with you to learn. I hope you can get something after reading this article. Let's take a look at it with the editor.

In linear regression analysis, independent variables (explanatory variables) are easy to be related to each other, which is called multicollinearity problem.

Moderate multicollinearity is not a problem, but when a serious collinearity problem occurs, it may lead to instability of the analysis results, and the symbol of the regression coefficient is completely opposite to the actual situation.

The independent variable that should be significant is not significant, but the independent variable that is not significant is significant. In this case, it is necessary to eliminate the influence of multicollinearity.

The cause of collinearity

The multicollinearity problem means that the change of one explanatory variable causes the change of another explanatory variable.

The original independent variables should be independent variables, so that according to the test results, we can know which factors have a significant impact on the dependent variable Y and which do not. If there is a strong linear relationship between the independent variables x, other variables cannot be fixed, and the real relationship between x and y cannot be found.

In addition, the reasons for multicollinearity may include:

Insufficient data. In some cases, collecting more data can solve the problem.

Use virtual variables incorrectly. (for example, if both male and female virtual variables are put into the model at the same time, collinearity must occur, which is called complete collinearity)

Discriminant index of collinearity

1.vif value

There are many methods to detect multicollinearity, and the more commonly used is the VIF value in regression analysis. The larger the VIF value is, the more serious the multicollinearity is. It is generally believed that the VIF is greater than 10:00 (strictly 5), which means there is a serious collinearity problem in the model.

two。 Tolerance value

Sometimes the tolerance value is used as the standard, and the tolerance value = 1/VIF, so if the tolerance value is greater than 0.1, it means there is no collinearity (strictly greater than 0.2). There is a logical correspondence between VIF and the tolerance value, and you can choose either of the two indicators.

3. Correlation coefficient

In addition, it is also a judgment method to directly analyze the correlation of independent variables and check the correlation coefficient and significance. If the correlation coefficient between one independent variable and other independent variables is significant, it means that there may be a multicollinearity problem.

Multiple collinear processing method

Multicollinearity is universal, usually if the collinearity is not serious (VIF stepwise regression)

3. Increase the sample size

Increasing the sample size is a way to explain the collinearity problem, but it may not be suitable in practice, because it takes cost and time to collect the sample size.

4. Ridge regression

The first and second solutions mentioned above are often used in practical research, but the problem is that if some independent variables do not want to be eliminated in the actual research, some independent variables are very important and can not be eliminated. At this time, only Ridge return may be the most suitable. Ridge regression is the most effective method to solve the collinearity problem at present.

Use path: advanced method > Ridge regression

Other instructions

1. Multicollinearity is common, mild multicollinearity problems can not be taken, if the VIF value is greater than 10 indicates that the collinearity is very serious, this situation needs to be dealt with, if the VIF value is below 5 does not need to be dealt with, if the VIF is between 5 and 10 depending on the situation.

two。 If the model is only used for prediction, the problem of multicollinearity can not be dealt with as long as the degree of fitting is good. When there is a multicollinearity model for prediction, it often does not affect the prediction results.

The above is how to solve big data's multicollinearity problem. The editor believes that there are some knowledge points that we may see or use in our daily work. I hope you can learn more from this article. For more details, please follow the industry information channel.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.