A deep learning approach to predict individual internet voting use based on electoral register data

The Swiss cantons are in charge of implementing all matters regarding referendum votes and elections, including the option to offer internet voting to their citizens residing in Switzerland or abroad. Geneva as one of the internet voting pioneers in Switzerland had internet voting trials on and off since 2003. However, due to security concerns and financial issues, internet voting in Switzerland in general is currently on hold. In this unique study based on electoral register data originating from 2008 to 2018, we classify individual voting channel adoption with the help of machine learning. A deep learning approach with categorical embeddings is implemented to forecast individual internet voting use. Results show that internet voting use can be predicted with 82 percent accuracy and that among the exogenous socio-economic variables studied, age seems to play a particularly crucial role.