In part one, we introduced a scenario – using a swarm of insects – to understand how to start planning for future outcomes. We introduced a very simplistic overview of a form of analysis called swarm intelligence. We also outlined the first two ingredients you’ll need in order to set up a foundation for reducing uncertainty and risks: 1) Fused Data, and 2) a High-Performance Analytics Engine.
For this post, as part two of our series, we get back to the original hypothetical problem you might have: limited resources to counter potential terrorist events in a geographic area of interest.
The pressing problem of predicting future mayhem – let’s use event risk forecasting.
For applying limited resources to counter potential terrorist events in a geographic area of interest, you might choose to apply behavioral theory and advanced analytics that others have used to reduce crime. One approach to try is known as the “near-repeat method.” This method uses the density of past events as a predictor of future event risk. Law enforcement might use this as a strategy to allocate policing resources, along with standard geospatial analysis tools to provide built-in support.
The near-repeat method of determining the risks of future violent events is straightforward: assign a high-to-low risk value to each past event. It’s simple because it requires only one data set (the violent events) without considering the underlying constraints and objectives of the attackers. However, this method isn’t as accurate, or what you will need, for complex operations.
Apply a Predictive Activity Model
Ingredient #3: Machine Learning and Artificial Intelligence
To get more accuracy than just using one dimension of crime stats, you can take a page from the swarm intelligence playbook: use an Event Risk Forecasting model to teach the machine (in your machine learning platform) of how past events actually happened. Just like you would personally try and understand what a terrorist or enemy might do next, you can train the machine learning model on a set of simple behavior rules (e.g., forests can hide the origin of bad guys, and roads can extend their resources). Then, add more rules and run parallel copies of your analysis (“swarming agents”).
Just as with the ant swarm scenario, using all the elements of behavioral data you fed the machine, the ERF engine traces the possible origins of terrorist attacks. From these findings, the artificial intelligence plays out future attack behavior, and creates an aggregated risk profile for future events.
To supercharge the process (and make it even more powerful), you can to run swarm analytics and swarm learning models continuously. This further reduces risk and improves your probability of better predictive outcomes by: (1) doing a better job of teaching your machine learning model with an understanding of known data and assumptions, and (2) giving you a process to adjust to newly-available data, or changes in assumptions by you or your analysts.
How can this help? Armed with all this data-crunching capability from the reporting of new events through intelligent data fusion, you are teaching the model to learn from the differences between what was forecast and what actually happened (through swarm learning). Future events and their locations (swarm forecast) are presented as a heat map [example above], showing the probabilities of occurrence, visually. With the outcomes presented, you can more confidently choose courses of action using some of that good old human intellect – which still isn’t a strong suit of our trained machines (yet).
Event Risk Forecasting in the Real World
A real-world example of tackling terrorist threats comes from the Armed Conflict Location & Event Data Project (ACLED). This is “a disaggregated conflict collection, analysis, and crisis mapping project. ACLED collects the dates, actors, types of violence, locations, and fatalities of all reported political violence and protest events across Africa, South Asia, South East Asia, and the Middle East.”
The folks at ACLED had the same challenge that you might face: How best to apply a limited number of security forces for a given location, while maximizing the ability to prevent terrorist attacks in a given geographic area. To do this, ACLED captured the forms, actors, dates, and locations of political violence and protests as they occurred across set geolocations.
ACLED then took a look at data related to political violence and protest events occurring within civil wars and periods of instability, public protest, and regime breakdown.
When we processed the ACLED data from 2012, using intelligent data fusion and high-speed processing for determining rebel attacks in Nigeria, the Event Risk Forecast model predicted 70% of subsequent events in 2013, focusing on only 10% of the geographic area.
While this model is applied to violent events in Africa, the same model would work for national security, defense or public-sector scenarios. Using an advanced approach with data fusion and predictive analytics, agencies can potentially better allocate personnel, protect critical infrastructure, improve or extend force protection, or make more specific, finite and effective deployment decisions, just to name a few.
Summary – a Three-Phased Approach to Better Predictions
Building this capability requires some maturity in the process. Offered below is a phased approach that can help folks get from very basic, foundational requirements, to capabilities that could provide better future outcomes for their operations.
Phase 1: Understand and leverage all your data
Understand where it lives, what is important, and how to access it. Develop a robust foundation for enterprise information management containing data integration, data quality, data profiling, and text analysis. Choose better courses of action from a data environment that is integrated, transformed, and able to deliver trusted data for your critical mission processes and decision makers through a timely, ingestible method. Faced with huge data volumes and disparate data sources, mission users must be able to harness and make sense out of chaos.
Phase 2: Develop your analytical insight
Establish a high-speed analytics platform that is able to handle the scale and scope of data – with tools for generating advanced analytics from your applications and sources. Once you’ve tackled phase one, your data platform will be able to provide natural language processing, geospatial, predictive, and graph capabilities. This will lead to you having an enterprise-wide intelligence and analytics platform that simplifies analysis across multiple, scattered, heterogeneous applications and platforms.
Phase 3: Apply a predictive activity model
Leverage the data and analytics that were liberated in phases 1–2. Machine learning will add immense value by infusing an entire set of solutions with more intelligence, at higher predictive rates, and on a platform for people (not just data scientists – whom we still think are awesome, btw). The entire ecosystem will be able to generate intelligence and increase mission capabilities.
Get in Touch
For more information on the Event Risk Forecast model, contact: firstname.lastname@example.org
For more information on High Performance Analytics, contact: email@example.com
 “Translating ‘Near Repeat’ Theory into a Geospatial Policing Strategy,” Police Foundation.
 “Repeat and Near Repeat Analysis,” ArcGIS for Government, Environmental Systems Research Institute Inc.