Australian GP 2017
Hereby I present you with the first predictions made by the algorithm. The model has not seen any data from the recent testing in Barcelona so everything is based on the drivers’ and teams’ performance in the past years. I expect the accuracy of the predictions to increase as the season progresses. Race predictions have been updated after the qualifying.
Personally, I believe the Ferraris will be well ahead of the Red Bulls. However, I’m not sure about how the battle between Mercedes and Ferrari will turn out. I also think that Bottas will qualify in the top 4 positions and I doubt Alonso can finish in a point-scoring position.
Driver | Qualifying Prediction | Actual Qualifying Result | Race Prediction | Actual Race Result |
---|---|---|---|---|
Vettel | 3 | 2 | 3 | 1 |
Hamilton | 1 | 1 | 1 | 2 |
Bottas | 7 | 3 | 2 | 3 |
Raikkonen | 5 | 4 | 4 | 4 |
Verstappen | 4 | 5 | 5 | 5 |
Massa | 8 | 7 | 6 | 6 |
Perez | 6 | 11 | 11 | 7 |
Sainz | 15 | 8 | 10 | 8 |
Kvyat | 12 | 9 | 9 | 9 |
Ocon | 13 | 14 | 13 | 10 |
Hulkenberg | 10 | 12 | 12 | 11 |
Giovinazzi | 20 | 16 | 18 | 12 |
Vandoorne | 16 | 18 | 19 | 13 |
Alonso | 9 | 13 | 14 | R |
Magnussen | 14 | 17 | 16 | R |
Stroll | 18 | 19 | 17 | R |
Ricciardo | 2 | 10 | 7 | R |
Ericsson | 17 | 15 | 15 | R |
Palmer | 19 | 20 | 20 | R |
Grosjean | 11 | 6 | 8 | R |
4 thoughts on “Australian GP 2017”
Did you decide to keep all other season testing and practice sessions out of the input dataset?
Any reason why you didn’t decide to include pre-season testing in Barcelona?
Yes, the input dataset only contains races and qualifying sessions; no practice sessions are used.
I mainly did it because I couldn’t find historical data for all practice sessions. My main source of data is http://ergast.com/mrd/ where data from practice sessions is not available.
For the same reason, I did not include data from the winter testing sessions.
Do you happen to know any such source of data?
One of the reasons why the teams run in FP2 is to feed their stochastic predictions for the race, with “real life” data.
Car & driver performance can very much be dependent of the track. Using season history can generate wrong outputs.
Although drivers behavior can be traced throughout the seasons, their pace is heavily dependent of the car they’re driving. The car pace is a variable that is renewed at least every year – in real life, this has to be adjusted on a race-to-race basis.
Totally agree with you! Especially this year, with so many regulation changes, the cars are not comparable to last year’s.
Therefore, I’m expecting the prediction accuracy to increase in the next few races i.e. after the model is fed with data that refer to this year’s cars.