Forecasting Significant Societal Events Using The Embers Streaming Predictive Analytics System

Area of Science:

• Computational Social Science
• Big Data Analytics
• Machine Learning

Background:

• Societal events like civil unrest pose significant challenges for prediction.
• Large-scale data analysis offers potential for early detection and forecasting.

Purpose of the Study:

• To describe the architecture and machine learning models of the EMBERS system.
• To evaluate the prospective forecasting performance of EMBERS for significant societal events.

Main Methods:

• Developed a big data analytics system (EMBERS) using a streaming, scalable, loosely coupled, shared-nothing architecture.
• Utilized ZeroMQ messaging and JSON data format, deployed on Amazon Web Services with automated deployment.
• Employed machine learning models for continuous, automated analysis of public data.

Main Results:

• EMBERS has been operational since November 2012, delivering approximately 50 daily predictions for Latin America.
• A detailed prospective evaluation of EMBERS' forecasting accuracy over the past two years was presented.

Conclusions:

• The EMBERS system provides a scalable big data analytics framework for forecasting significant societal events.
• The system's architecture and machine learning models demonstrate effectiveness in real-world event prediction.