BENV0091 Group Project: Predicting Arctic Sea Ice with Supervised Learning
Media: on UCL Energy Institute's post
Visualisation: Shiny App
Report in PDF: https://realgjl.com/report/ucl/arctic_sea_ice.pdf
We are heavily using Jupyter Notebook as our developing platform to implement, build, and test our SL models for environmental risks on Arctic sea ice, and using miniconda to create isolated environments and manage our R packages.
For new/existing users of miniconda/Anaconda, please go to Install miniconda & set base environment and/or For existing conda users: create a new and isolated R environment.
tidyverse corrplot MASS randomforest neuralnet
conda install -c conda-forge r-tidyverse r-corrplot r-mass r-randomforest r-neuralnet
- Arctic sea ice: National Snow & Ice Data Center
- CO2: National Oceanic and Atmospheric Administration
- Ozone: NASA Goddard Space Flight Center
- Temperature: National Oceanic and Atmospheric Administration
- Rainfall and daylight: Weather US
- Population:Our World in Data
- GDP in current USD: World Bank & IMF (World GDP will fall -4.4% in 2020, +5.2% in 2021, +4.2% in 2022)
| Model | R-squared | MSE |
|---|---|---|
| Linear Regression | 0.898 | 0.00946 |
| Penalized Linear Regression (Lasso, min) | 0.863 | 0.00690 |
| Penalized Linear Regression (Lasso, 1se) | 0.857 | 0.00734 |
| Penalized Polynomial Regression (Lasso, min) | 0.931 | 0.00448 |
| Penalized Polynomial Regression (Lasso, 1se) | 0.917 | 0.00483 |
| Random Forest | 0.983 | 0.00105 |
| Neural Networks | 0.927 | 0.00106 |