How to prevent google adsense from spoofing click analysis

With the popularity of online advertising, the pay by per click (pay by each click) model is gradually accepted. However, the problem that follows is fraud.
The prevention of clicking is at ease, because it is directly related to whether this advertising model can survive for a long time and become a real source of income for website owners.
The following describes how the Google Adsense system can prevent click spoofing from the system perspective, hoping to provide guidance for other online advertising systems to prevent fake clicks:
1] clicks = clicks/total views.
Click rate is a key method to determine whether there is any fraud clicks. It can be imagined that the click rate of an advertisement on a website is over 10%.
# Of Click/# of Viewed
2] Click coverage/independent IP address, if any in this distribution; single IP address (click/browse) = click coverage exceeds 3 times of the system error range will be suspected of cheating.
For example, if a user from 129.119.200.1 browses 16 webpages and clicks 4 advertisements, the click rate of the entire advertisement "calculated from [1]" is 5%, the calculation result is as follows:
% 5X16 = ~ 1. The variance is Sqrt (1) = 1, and the click coverage rate is = 4/1 = 4. According to the mathematical Gaussian distribution, this probability is less than one in ten.
Ratio vs ip distribution
3] click rate "Click coverage"/IP/time
Analysis of click-through rate based on time series. If there is a significant peak value in a certain period of time, this may be a potential possibility of spoofing clicks.
Ratio VS time
4] Analysis of webpage load time and ad click time difference, and analysis of time difference sequence between each two clicks
[Time difference between webpage load and ad click] it should be a Poisson distribution possion
Distribution, and the time difference between two clicks should also be a Possion
Distribution. If this time is recorded in seconds, the Gaussian distribution is basically displayed if the value is greater than 25 seconds.
[Time of loading-time of click] distribution VS Possion
[Time difference of two clicks] distribution VS Possion/Gaussion
5] Proxy click analysis
Changing the IP address to click can be said to be the most difficult to solve in the past, the most difficult way to find cheating, probably Chinese people in the Alexa Boost most of the use of Proxy for false click method, however, you only need to check whether the IP source is a server with the Proxy function through reverse monitoring.
Reverse Proxy check
6] http_agent analysis
Http_agent/time series analysis, peak value exceeds 3 variance needs to be reviewed
7] Analysis of http_referral
Referral/time series analysis, peak value over 3 variance needs to be reviewed
8] The overall effect is also very useful: