Graph CNN for prediction of voting behaviour with census data
Hierachical Graph Convolutional Neural Network able to integrate data from multiple graphs using embedding and projection operations in a residual approach. Based on spatial graph convolutions.
Applied to Census data to predict election outcomes - provides significant improvement in RMSE - a 3 fold decrease compared to generalised linear methods.
See paper at https://www.aaai.org/Papers/AAAI/2019/AAAI-LiM.4653.pdf