Improving situational awareness for humanitarian logistics through predictive modeling

Abstract

Humanitarian aid efforts in response to natural and man-made disasters often involve complicated logistical challenges. Problems such as communication failures, damaged infrastructure, violence, looting, and corrupt officials are examples of obstacles that aid organizations face. The inability to plan relief operations during disaster situations leads to greater human suffering and wasted resources. Our team used the Global Database of Events, Location, and Tone (GDELT), a machine-coded database of international events, for all of the models described in this paper. We produced a range of predictive models for the occurrence of violence in Sudan, including time series, general logistic regression, and random forest models using both R and Apache Mahout. We also undertook a validation of the data within GDELT to confirm the event, actor, and location fields according to specific, pre-determined criteria. Our team found that, on average, 81.2 percent of the event codes in the database accurately reflected the nature of the articles. The best regression models had a mean square error (MSE) of 316.6 and the area under the receiver operating characteristic curve (AUC) was 0.868. The final random forest models had a MSE of 339.6 and AUC of 0.861. Using Mahout did not provide any significant advantages over R in the creation of these models.