Google guess on the basis of the anti-cheating

1] CTR = clicks / total number of the browser.
CTR is a fraud clicks to determine whether any of the key ways we can imagine a Web site advertising the hits more than 10% of what it will mean.
# of click / # of viewed

2] Click on the coverage / independence ip, in the distribution if there is; of a single ip (click / view) = click coverage beyond the 3-fold margin of error of the system, there would be suspected of cheating.
For example, 129.119.200.1 from the user's browsing 16 pages, click on the ads 4, and the entire advertising hits "from the [1] in the calculation of the" 5%, calculated:% 5 x 16 ~ = 1, the variance for sqrt (1) = 1, click coverage = 4 / 4 = 1, according to the Gaussian distribution mathematically, the probability is less than one ten thousandth.
ratio vs ip distribution nd

3] hits' click 'coverage / ip / Time
According to the hits of the time series analysis, if at a certain time period on a clear peak, it will be that there is a potential click fraud possible.
ratio vs time o

4] page load time and ad clicks of the time difference, as well as every time difference between the two click sequence analysis of the fish you irresolute nuclear o?
[Page load time and the time difference click advertising] is a Poisson distribution possion distribution, and each click between the two is also the time difference should be a possion distribution, if the second time in mind that more than 25 seconds, then basically a Gaussian shape.
[time of loading - time of click] distribution vs possion 4 $
[time difference of two clicks] distribution vs possion / gaussion
Jihui999.cn reproduced from the content of this reference, please!
5] for proxy
Ip change for clicks can be said that in the past, is the most difficult to resolve the most difficult to find ways to cheat, people probably alexa for the boost on most of the use of proxy for the false-click method, but through here as long as the reverse of the Board of Audit and Inspection ip source is to bring There are features of the proxy server can see.
reverse proxy check

6] for the analysis of http_agent
http_agent / hours of time series analysis of peak need to review more than 3 variance

7] of the http_referral for analysis
referral / hours of time series analysis of peak need to review more than 3 variance

8] on the overall results have a very useful volume:
All the user's effective CPM of the mean / ip independence
This will be able to find a more direct spam clicking the computer running and be blocked.