Risk Assessment of Deadly Economic Socio-Political Crisis with Correlational Network and Convolutional Neural Network

From social analysis to biology to machine learning, graphs naturally occur in a wide range of applications. In contrast to studying data one at a time, graphs' unique capacity to capture structural relationships among data enables them to yield additional insights. Nevertheless, the capacity to learn from graphs can be difficult because meaningful connectivity should exist between data and the form of data such as text, numbers or categories should allow for building a graph from their relationships. Investigating hidden patterns in the variation of development indicators and severe socio-political crises that happened in low-income countries is an analytical approach that has been experimented with in this research. Evidence of a correlation between socio-political crises and development indicators suggests that a method to assess the risk of crisis should consider the context of each country, as well as the relative means of crisis. This research reviewed different risk assessment methods and proposed a novel method based on a weighted correlation network, and convolution neural network, to generate images representing the signature of development indicators correlating with a crisis. The convolution neural network trained to identify changes in indicators will be able to find countries with similar signatures and provide insights about important indicators that might reduce the number of deadly crises in a country. This research enhances the knowledge of developing a quantitative risk assessment for crisis prevention with development indicators.