I have a dataset containing a hobby project of mine: http://pastebin.com/y540VP1U It is a table (space separated)

id  track         driver         vehicle   km_h   standard_vehicle_km_h
L25 LosSantosGP   StarlaLeyba    Verlierer 134.40 142.01
L26 LosSantosGP   YolandeApodaca Verlierer 134.24 142.01
L27 LosSantosGP   TeenaCuenca    Omnis     126.36 137.93
L17 NODOGrapeseed StarlaLeyba    Fusilade  157.55 135.81
L18 NODOGrapeseed JoellenSeneca  9F        166.07 145.40

containing these fields:

  • id - Unique identifier for the row
  • track - The name of the racetrack driven on
  • driver - The name of the person who drove on the track
  • vehicle - The name of the vehicle the driver used
  • km_h - The average speed (kilometers/hour) of this driver in this vehicle on this track.
  • standard_vehicle_km_h - The average speed of one specific person (let's call him Broughy) on a reference track (namely CuttingCoronersGP) in the vehicle

Broughy drove every vehicle available on the reference track CuttingCoronersGP, and recorded the average speed for each vehicle. I am certain, that he is a) a very good driver, and b) his recorded speeds are consistent. In the dataset you'll also see (e.g. lines 65–91) that some people drove on the reference track "CuttingCoronersGP" as well.

What I am trying to find out is:

Predict the average speed of a person when he announces to drive vehicle X on track Y with reasonable confidence.

My thoughts until now (I'm pretty much a statistics layman, but an ok software developer):

I need a multiplier t for the different tracks, since they can vary greatly in speed (lots of tight corners versus an oval "Indianapolis"-like track). I would need a couple of recordings of drivers to determine how much slower or faster the track is in comparison to the reference track CuttingCoronersGP.

I also need a multiplier d for each driver, since some drivers are really fast, and some just began driving. I might need about 10 recordings of a person to be able to predict his multiplier. I may use the median of his recorded speeds (after discarding outliers).

Finally to predict the speed any driver would take on any given track in any vehicle would be standard_vehicle_km_h * t * d

But since d and t seem pretty dependent on each other, I'm stuck.

Maybe someone can have a look at the dataset, play around a bit with it and has an idea how I am able to do a meaningful prediction for any driver on any track?

  • $\begingroup$ This looks like a standard regression problem unless I misunderstand you. Have you considered that? $\endgroup$ – mdewey Mar 20 '17 at 18:09
  • $\begingroup$ Since I'm a layman, I didn't. But I will try to read stuff on the topic. $\endgroup$ – mawimawi Mar 21 '17 at 7:27

Sounds like you probably want a regression model, where the dependent variable is average speed and the independent variables are track, driver, and vehicle. (standard_vehicle_km_h is of limited use without a much more detailed model of average speeds.) The simplest kinds of regression models should be discussed in any introductory text on data analysis, statistics, or machine learning, which you should avail yourself of if you want to analyze data.

  • $\begingroup$ I skip-read a couple of articles online, and all the independent variables were numeric (which makes sense). In my case, I have labels. E.g. "JosefineLipka" as the name of a driver. For tracks I could use track_length in km, but I also have no idea which value to use for vehicle. I am sorry to maybe annoy you here, but I'm stuck here when I try to find sensible numeric values. $\endgroup$ – mawimawi Mar 21 '17 at 7:57
  • $\begingroup$ Just realized: track_length has no meaning as well. $\endgroup$ – mawimawi Mar 21 '17 at 8:02
  • $\begingroup$ Please ignore my comments. Found something about "categorical" predictors... reading... Sorry again for spamming here, just wanted to let you know that I might be on the right track myself now. $\endgroup$ – mawimawi Mar 21 '17 at 8:46
  • 1
    $\begingroup$ @mawimawi You really ought to work through basic statistical concepts first, before tackling real problems. Statistics isn't something you can do well after a one-day crash course. A lot of bad data analyses are done by computer programmers who think that statistics takes no more investment than learning a new programming language. $\endgroup$ – Kodiologist Mar 21 '17 at 16:12
  • $\begingroup$ yes, I realized that yesterday... Anyways, I'm thankful for the "regression" hint, and I think I'm starting to grasp the concept behind it. It's really interesting, and I might dig deeper into statistics if/when I have more free time, since I only need it for this hobby project at the moment. $\endgroup$ – mawimawi Mar 22 '17 at 8:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.