Common prediction methods and features

Common prediction methods and features

Qualitative and quantitative	Method Name	Applicable time	Method description	Data required
Qualitative Method	Expert Conference Law	Long-term prediction	Organize experts to make predictions in the form of meetings, and draw conclusions based on experts' opinions.	Historical Market Development Information
	Delphi Method	Long-term prediction	Experts will develop in accordance with the law, conduct anonymous surveys on multiple experts, conduct statistical analysis on the results through multiple rounds of feedback, and generate quantitative results using average or median.	Comprehensive Analysis of expert opinions
	Analogy Prediction Method	Long-term prediction	Compares and analyzes the appearance and development of similar products based on the principle of development similarity of things.	Related Historical Data
Quantitative Method	Linear regression prediction (including mona1 and multivariate)	Short and mid-term prediction	There is a linear relationship between the dependent variable and one or more independent variables. The most common one-dimensional model relationship is y = ax + B.	Historical data of the volume to be analyzed
	Nonlinear regression prediction	Short and mid-term prediction	There is a non-linear relationship between a dependent variable and one or more independent variables. Common relational models include power function, exponential function, parabolic function, logarithm function, and s-form function.	Historical data of the object to be analyzed
	Trend Extension Method	Medium-to long-term prediction	Use mathematical models to fit a trend line and predict the future development trend of things	Long-term Historical Data
	Average moving Method	Short-Term Prediction	Use the average value of multiple data periods to predict future trends	Multiple historical periods.
	Exponential Smoothing	Short-Term Prediction	Similar to the mean moving method, but different weights are given to recent and distant observations to predict trends.	Multiple historical periods.
	Adaptive Filtering	Short-Term Prediction	The nature of the trend pattern changes with time, and there is no seasonal time series data	Multiple historical periods.
	Smoothing time series prediction method	Short-Term Prediction	Prediction methods applicable to the development forms of various sequences	Multiple historical periods.
	Intervention analysis Prediction Model	Short-Term Prediction	A Prediction Method Applicable to time series when a node is affected by unexpected events.	Historical data and major impact events
	Boom Prediction Method	Short and mid-term prediction	Continuation and turning prediction of time series Trends	Large amount of historical data
	Grey Prediction Method	Short and mid-term prediction	At that time, the development was exponential.	Analyze historical data of an object
	State space model and Kalman Filter	Short and mid-term prediction	Suitable for prediction of various time series	Analyze historical data of an object

Introduction to qualitative methods and specific cases

1. Expert Conference method and Delphi Method

Both of these methods rely on the judgment of experts to make predictions. They are subjective. Therefore, its effectiveness is very limited to the professional level of experts, so its practicality and reliability are relatively poor.

The basic process of expert meetings is to organize relevant materials> organize expert meetings> give experts their own ideas> summarize the results.

The Delphi method is different in that there is no face-to-face meeting process, and experts make their own judgments anonymously.

There are three methods to summarize the results:

A. Three-point estimation method. That is, the participating experts give three possible values: highest, most likely, and lowest.

B. Relative Importance method. Based on the knowledge and experience of the experts, divide them into different weights and calculate the weighted average for the prediction results based on the weights.

C. subjective probability method. The forecaster makes a subjective judgment on the probability of the event to be predicted, and gives the possible probability.

In practical application, the above methods may be used for interaction. The example in the reference document is the combination of the three-point estimation method and the relative importance method. For example, Manager, b, c, very easy to understand. A good example of the subjective probability method is Chapter 7 subjective probability in "in-depth analysis of data", which also involves how to use Bayesian rules to correct subjective probability, if you can find this book online, you can check it out.

Reference: http://www.docin.com/p-648680864.html

Http://wenku.baidu.com/link? Url = GRHjufflZW3vgV-GOBncpYhSRSyeHG7NEkJ16zJnQMjDWiCCopYaAwXRBluIapJpKAL6ewrjICyi0KJJ9y7MjfHdtzk2dfi_uh2SdqfelGC

 

2. analogy Prediction

Analogy prediction is based on the differences between the comparison objects, such as product analogy, Regional Analogy, International analogy, Industry Promotion, and updating. Predict the possible market performance of a new product based on the Market Performance of a known product, or predict the possible market performance of a product in region. However, the similarity of events is limited, and different events are affected by many different factors. Therefore, analogy prediction is not a strict prediction method, and the effect is limited. However, it has many applications. The second reference Link provides an example of Regional Analogy prediction.

Reference: http://wenku.baidu.com/view/e4d3a1c50975f46527d3e175.html

Http://wenku.baidu.com/view/7f60811ceefdc8d376ee326a.html

 

Reference Book: Statistical prediction and Decision-Making edited by Xu Guoxiang

In-Depth Data Analysis

Reference: http://wenku.baidu.com/view/53f30835a32d7375a41780d3.html

The quantitative prediction method will be further studied and summarized.

If you have any questions or suggestions, please submit them. Please indicate the source for reprinting. Thank you!

Common prediction methods and features