-
-
Notifications
You must be signed in to change notification settings - Fork 1k
GSoC_2017_applications_elections
2016 saw some spectacular popular votes, such as the UK “Brexit” referendum and the US presidential elections. Importantly, the outcomes of these votes were deemed surprising. For example, in the election of Donald Trump as 45th president of the United States major news coverage - up until very late during election night - predicted a win for Hillary Clinton. Can it be so difficult to predict understand and predict voters’s behaviour - even if there are only 2 options (when on the other hand whole personalities can be reconstructed from someone’s facebook preferences?
- Easy to Medium
- data wrangling/ data analysis skills
- some Shogun and Machine Learning basics
- some statistics and background knowledge on elections/demographic analysis is helpful
Europe has a year of crucial elections ahead of itself, for example France presidential elections, Germany parliamentary elections. The challenge of this GSoC Data Project will be to use Shogun tools in addition to classical data analysis of voter characteristics and understand the election results. Depending on your interest, we can put the focus on mapping voter characteristics or predicting/explaining election outcomes (see for example the Digital Times analysis of the UK 2015 elections which explains their method and even provides code).
- Identify datasets on voter demographics, such as census data, GIP, education levels, as well as surveys on voting intentions for both previous elections and the most current data
- Clean data and map voter features, e.g. heat map with Shogun's Kernel density estimation and/or spatial modelling with Shogun's Gaussian processes
- Find predictors for voting behaviour in each constituency (Shogun's dimensionality reduction might be useful here)
- Shogun's Decision Trees to learn predictors
- Predict an upcoming election/ explain a previous election
- Document the process and show the results - blog entries, a sleek (interactive) visualisation
With a real world data project we can - finally - use the power of Shogun on data that really matter. In particular, understanding voter behaviour is not just interesting and of immense political importance; predicting election outcomes is also a notoriously difficult challenge - and the actual events have a huge impact on a lot of people's lifes. So making the right prediction or providing a good explanation for the result really matters!
- Predicted wins for Clinton from CNN and FiveThirtyEight
- Digital Times Analysis of UK 2015 elections with their code
- UK 2015 Results Analysis
- Voting behaviour predictors decision tree
- Decision trees in Shogun