Is there any work already made with predicting number of visits by POI? What models of machine learning would you recommend for this goal (predicting daily number of visits by POI)
Hi @Angel_Langdon, I’m not aware of any machine learning models that have been used to predict the daily number of visits to a POI, but have you seen the Data Science Resources? There are a couple resources that describe methods for estimating true counts from SafeGraph’s data. The best approach right now is “CBG micro-normalization”, where the raw visit counts are adjusted based on the # of SafeGraph devices and US Census population in the visitor home CBGs.
The difficult part of building a machine learning model is that true counts data is hard to come by. Another challenge is that predicting true visits for a specific POI can be noisy; predicting for an entire brand tends to yield more reliable results.
However, we are constantly trying to improve the methods!
Thank you very much! We will apply the CBG micro-normalization, and later when we have all our data features will try to predict the number of visits.
You’re welcome! Let me know how it goes!
Hi!, We have applied CBG micro-normalization and with a simple regression model we have obtained 0.65 pearson coefficient, which is pretty good considering that we are using a simple regression model…
Hi @Angel_Langdon, that is pretty solid! What types of locations are you using? Do you have true counts for the predictions?
We are using patterns data from Subways in Houston between 2020-03 until now.
Sorry, I am from Spain thus I am not really really good at English, what do you mean by true counts for the predictions?
@Angel_Langdon, no worries, I think your English is excellent! I was just wondering if you are working with locations that have “true counts” data available. For example, airports publish the number of people who get on a plane each day. With the true counts of visitors, you are able to build a model that more accurately predicts the number of visitors from SafeGraph data. I was just curious if you had that data or not. Sorry for the confusion!