We are using historical F1 data from 1950 - 2019 in order to create a prediction for the 2020 season. We will then compare this to the results of the actual 2020 season.
Difficulties:
- There have been many changes over the decades in the F1 sport.
- Qualifying times and rounds need to be normalized and properly collected.
- There is also a lot of missing data from earlier decades and extra data from recent years.
- Use the Ergast F1 API to collect data (this is what most Kaggle datasets use)
- Store this data in a sqlite database (we also use sqlalchemy to better interact with our sqlite engine)
- Clean, merge, and feature engineer our collected data from the database, and store in a cleaned table
- Split the cleaned data 80/20. Use the training set to traing our Forest Regressor Model. Test with testing set
- Use our model to predict the 2020 season results
We're using a requirements.txt file to specify all the external libraries and their versions that our project needs to run. We are also using a virtual environment to run the modules.
-
start a virtual environment with
- python3 -m venv venv
- source venv/bin/activate
- pip install -r requirements.txt
-
Alternatively all the modules from requirements.txt can be pip installed individually
-
test database connection and data collection
- python3 test_collections.py
Overall by running this project from 2015 to 2020 (training from 2015-2019), our model was able to correctly predict the top 4 out of 5 positions for the 2020 season.
Note: This submission only contains the PDF because our submission is too large. Everything else is in the provided GitHub Repository. This is also the link to the files on Google Drive: https://drive.google.com/drive/folders/1mEIl1QKqKNLVRyPchi0SlkhFKDHZ_gXI?usp=sharing