Using artificial intelligence, it’s possible to identify patterns in nearly any data set large enough. In this case study, we at ConvergentAI™ set out to identify the attack patterns of the terrorist group called Boko Haram in Nigeria. The ISIS-affiliated terrorist group is said to have been formed in the 1990s, but attacks have been ramping up pretty rapidly since early 2011. With the increase in attacks by Boko Haram on military bases, churches, and towns, Nigeria needed to improve how they dispersed their limited security forces to mitigate future attacks.
This type of challenge is no stranger to artificial intelligence. Behavioral theory and advanced analytics have been at the forefront of reducing crime for quite some time, but the current state of the art method, Near Repeat, needs to be much more accurate. This method of determining the risk of future violent events is pretty straightforward and not very intelligent. In its simplest form, it assigns a high-to-low risk value in a circle around each past event in a data set. It’s simple because it only requires one data set (the violence events) and only assigns inherent risk around the event. It has proved to work well in the past, but when lives are on the line, any improvement to accuracy and speed is very valuable.
With the pitfalls of the Near Repeat method and the severity of the Boko Haram attacks in mind, we set up a plan to create a system that could forecast future high-risk areas quickly and accurately. We needed the “agents” that set the risk values on the topology to be flexible, smart, and be able to adjust to changing data. Using a dataset of past events, we created “origin agents” that worked backwards in time to try to determine the areas with the highest chance of containing Boko Haram bases of operations. Once the system felt comfortable with the results of the possible origin locations, these locations were used as a starting point for all of the “advanced risk agents”. These agents work by accounting for many different variables (landscape, weather, roadways, terror event type, etc.) and intelligently using this data to more accurately forecast risk for possible future attacks in a much shorter amount of time.
Our Streaming Predictive Analytics was created not only as a way to utilize and test our advanced method of forecasting risk, but also as a way to demonstrate how our method compares to the current baseline, Near Repeat. Here is our most used test case and the results of both tests: Using the past events to test both methods, we tested how many violence events we could defend against by properly assessing which 10% of land in Nigeria was at highest risk for an attack. With the Near Repeat method, we were able to forecast and defend against 45% of the attacks while defending 10% of the land. With our advanced risk forecasting method, we were able to forecast and defend against 70% of the attacks while defending 10% of the land. We concluded that by using our method we were able to prepare and defend against 55% more attacks than the Near Repeat method.