Malicious behavior has become digital: Crimes like credit card fraud & identity theft occur at an unparalleled scale and frequency. Detecting these acts is crucial - but humans are no longer able to keep up with the speed and volume.
Based on our industry experience we built a prescriptive AI to predict & counter crimes in the city of Chicago. The case illustrates the problem of automated resource allocation when dealing with threats.
We share our modelling approach and show an interactive working system based on real live data.
You will get to know our heuristics for tool-selections and how you can transition from aggregated visuals to your own sophisticated machine learning approach.
Target Audience: Everyone interested in AI, or predicting behavior. Especially people from the cyber risk or risk management field
'How can AI systems support humans in risk management?
A critical part of risk management is detecting malicious behavior. We instinctively perform threat assessments by monitoring the behavior of other humans. However, these instincts do not help us in today's digitized world. Can AI help us to discover malicious intent in digital behavior?
This malicious behavior may come in many forms: credit card fraud, identity theft, ransomware etc. The side effects of the digitization do not only include the emergence of new species of malicious acts but also the frequency and scale at which they occur. Detecting these acts has become a crucial activity for any digitized business, such as the financial industry in which the stakes are especially high.
Tackling the task of detecting malicious behavior with humans, who aggregate and visualize data in BI systems to make choices and formulate actions, is not enough:
Humans are biased, limited in their cognitive capacity, very expensive and despite their shortcomings even demand paid vacation.
The industry has reacted to the increased demand for decision making and is now extending their BI products with AI capabilities aiming for decision support. Many products are marketed as all-purpose AI solutions. At the same time, many open source toolkits allow the user to build special purpose AI solutions. In this showcase we will answer the question: Should we strive to build or buy the best all-purpose AI or should we strive to find the best AI-toolkit?
We have been working on multiple projects located in the financial industry aiming to detect malicious activities and want to share our insights with you. We will give you an overview of our findings from a recent project where we identify risky behavior from IT-ticket data.
Since customer data is highly confidential, it is tough to make open-source showcases from real projects. We still want to show a real-world application to our audience, so we built a prescriptive model for malicious human behavior based on public data.
We will address the question if AI can help to prevent actual (physical) crimes in the city of Chicago, a city infamous for its high crime rate. To this end we utilize crime data made available by the Chicago police department and enriched by publicly available weather data.
We show how, by making predictions about future crimes, we can improve the utilization of police resources. This objective is analogous to many business scenarios: It is not enough to make accurate predictions - we need to translate these predictions into problem solving strategies. By doing this we transition from predictive to prescriptive models.
In the live demo we use Splunk as our BI software and Python as our AI backend. We visualize the historic crime data and identify hotspots of crime activity that change over the course of days. It is our goal to identify the hotspots of tomorrow. You will see, how we transition our decision making from aggregated visuals to a more sophisticated machine learning approach.
In many real-world projects, users are faced with a veritable zoo of possible tools, applications and databases. We will give participants insights into our heuristics for tool selection and our top picks for analytic toolsets. By the end of the talk you will have ideas in mind to answer questions like:
How to best utilize my available data?
How do I select the right toolkit or out-of the box solution?
How to transition from simple to sophisticated AI models?
How to structure AI projects in the real world?