Ranking a Location using Linear Regression

[Confidential]

Hi,

I am facing a strange problem with a software patented in the USA. It is a location evaluation software, which offers insights on determining the best location for starting-up a gas station / petrol station.

While reading the patent thoroughly, I feel that the software operates based on a complete gimmick and the realistic estimate it aims to deliver cannot be estimated in the way it is written. The software shall rank a lot of established gas stations in an area (training locations) and the proposed location. The ranks are generated through weighted sum of known data about the locations. The weights are generated by linear regression analysis or Pearson correlation of some unknown variables! At the end of the section explaining the statistical procedures, the patent reads, "In this Way, it has been found that a more accurate estimate of a value can be determined for the location based upon the training site data." This accurate estimate may refer to the future sales projection of the site, meaning that the forecast matches actual sales.

I would be happy if forum members can give their opinions about my understanding that the software is nothing but a Mambo Jumbo and the estimate it aims to provide through linear regression is a spurious estimate. The ranking of locations, which follows the regression analysis, is perhaps just based on traffic count at different locations. Your opinions shall help me to understand the merits of using this software and help me make a decision. The patent document is attached to the post.

Kindest regards,

Confidential

 

[statsanon]

Hi,

I am facing a strange problem with a software patented in the USA. It is a location evaluation software, which offers insights on determining the best location for starting-up a gas station / petrol station.

While reading the patent thoroughly, I feel that the software operates based on a complete gimmick and the realistic estimate it aims to deliver cannot be estimated in the way it is written. The software shall rank a lot of established gas stations in an area (training locations) and the proposed location. The ranks are generated through weighted sum of known data about the locations. The weights are generated by linear regression analysis or Pearson correlation of some unknown variables! At the end of the section explaining the statistical procedures, the patent reads, "In this Way, it has been found that a more accurate estimate of a value can be determined for the location based upon the training site data." This accurate estimate may refer to the future sales projection of the site, meaning that the forecast matches actual sales.

I would be happy if forum members can give their opinions about my understanding that the software is nothing but a Mambo Jumbo and the estimate it aims to provide through linear regression is a spurious estimate. The ranking of locations, which follows the regression analysis, is perhaps just based on traffic count at different locations. Your opinions shall help me to understand the merits of using this software and help me make a decision. The patent document is attached to the post.

Kindest regards,

Confidential

My initial impression is, without any knowledge or evaluation of the software is that you should generally avoid any evaluation that doesn't tell you the methodology behind that evaluation. There is a difference between proprietary and hiding important information. As someone with more than 20 years consulting experience, most in the spatial area, it was important to me and my clients that everything I did and supplied to my client was open to them. I would certainly not trust any software that claims what you say but does not inform you of the details of the variables and methodology behind it.

 

[statsanon]

Hi

Having read the application sadly I think it is mostly mumbo jumbo designed to try and impress a patent office. Sadly there is a long history of this where patent officers who are ignorant of the specific science/technology are tricked. And sadly the applicants are often given their patents for ideas over which they have no rights. The first few pages are unbelievable and wouldn't even trick most high school students.

PS If you wonder about my credentials I have over 20 years working in location base services and nothing there is really original and in my view it would be totally immoral for anyone to claim a patent over things that people do in their work day to day. Sadly though it hasn't stopped such patents in the past

 

[Confidential]

Thank you Statsanon for confirming my suspicion as legitimate. Since the company is advertising accurate predictions based on spurious regression forecasting, it is misleading customers. However, before contacting trading standard office, I would need an explanation in statistical terms explaining what is wrong in the model. This is where I need some suggestions perhaps. I expect they can make a counter-argument saying the model is right to deliver the prediction and the trading standard officer will get confused.

 

[statsanon]

Thank you Statsanon for confirming my suspicion as legitimate. Since the company is advertising accurate predictions based on spurious regression forecasting, it is misleading customers. However, before contacting trading standard office, I would need an explanation in statistical terms explaining what is wrong in the model. This is where I need some suggestions perhaps. I expect they can make a counter-argument saying the model is right to deliver the prediction and the trading standard officer will get confused.

hi
I didn't say there was anything wrong with the model per se. I was making a different point about what I saw as the lack of merit in the application based on originality and other factors.

 

[rogojel]

Hi,
maybe you could pick a random sample of known gas stations, let the SW evaluate them and compare the guess to reality?
regards

 

[Confidential]

Hi,
maybe you could pick a random sample of known gas stations, let the SW evaluate them and compare the guess to reality?
regards

Thank you @rogojel for your idea. I have spoken with a few IP experts and they think software IP is a grey area. If a statistical equation improves the efficiency of solving a problem, then it can be patented. Because regression analysis cannot be patented, one can try to put some other equations around it and claim an invention. This kind of statistical technique may not be put forward in academic journals and so we do not see any data analysis after the propositions made. This patent is just about a proposition that is not tested and the results are not validated. The trading standard authorities or advertising authorities can look at the business and say the company cannot claim accurate prediction in their advertisement contents. To challenge the patent at USPTO, a lengthy re-examination procedure needs to be launched. What I have learned from this experience is that I can package regression analysis with some other addition or multiplication of variables to claim an invention in business intelligence. Then, anyone else may not be able to use regression analysis.

Imagine if someone patents the procedure of solving basic math problems like how much time it will take to fill up a bucket with two taps. There is a step by step procedure to solve this math problem and if that procedure is patented, then others cannot use it. Amazing!