Parking Prediction based on Mobile application

Question

I am quite new to predictive modelling but have knowledge of GIS, R, python, SQL, etc.

I am currently doing a project in work trying to predict when parking spaces will be available based on data received from a mobile phone application.

I have 2 sql tables

ParkingTickets

ParkingTicketId  (Integer)
ParkingAreaId    (Integer)
Latitude (Integer)
Longitude    (Integer)
Location (Text)
ParkingDate  (DateTime)
DurationInMinutes    (Integer)
ExpiryDate   (DateTime)
Day (Text)

ParkingAreas:

ParkingAreaId (Integer)
MaxSpaces (Integer)

Main assumption is that only app users can park in the 18 parking lots (both on-street and off-street lots (no multi store or underground)). I do not take in to account user preference, weather, events, etc., etc. This is purely presence and absence based on data at hand. I have researched techniques and was looking into using a birth/death model but struggling to find examples of it in use.

Any help or pointers on models to use would be brilliant!

sample data:

ParkingTicketId    ParkingAreaId   Latitude    Longitude   Date    DurationInMinutes   ExpiryDate  Day
60465   302 42.56246869 -70.91313754    2014/03/07 16:36    5   2014/03/07 16:41    Friday
60466   302 42.57139883 -70.91906364    2014/03/07 16:36    23  2014/03/07 16:59    Friday
60467   302 42.54419925 -70.9417496 2014/03/07 16:36    24  2014/03/07 17:00    Friday
60472   302 42.57576595 -70.92876607    2014/03/07 16:36    16  2014/03/07 16:52    Friday
60477   302 42.55573294 -70.92912634    2014/03/07 16:36    9   2014/03/07 16:45    Friday
60479   302 42.55711998 -70.91200458    2014/03/07 16:36    19  2014/03/07 16:55    Friday
60480   302 42.58008043 -70.91559081    2014/03/07 16:37    5   2014/03/07 16:42    Friday
60485   302 42.55161223 -70.9240808 2014/03/07 16:37    21  2014/03/07 16:58    Friday
60492   302 42.58437849 -70.92764527    2014/03/07 16:37    6   2014/03/07 16:43    Friday

ParkingAreaId   MaxSpaces
302 8
304 50
306 95
308 30

Ryan Zotti · Answer

If this data is coming in real-time then you don't need a model -- simply check how many spots have an ExpiryDate greater than now (i.e. when you need to provide a prediction) and subtract this from the total capacity of spots.
If the data is not coming in real-time, then you could use time of day and day of week as predictors. You might even want to make them into interaction terms. You would also need to decide how often you want to call your model and group your data so that each row represents how many tickets were active during that timeframe; this will define your target variable (what you're trying to predict).
By the way, I think you're referring to a survival model. I would recommend gradient boosting instead; it's much more powerful. Gradient boosting models (GBM) are part of caret in R and scikit-learn in Python by the way.

Parking Prediction based on Mobile application

One Answer

Add your own answers!

Ask a Question