Diep Dao has co-developed a new analytical framework for crime research that could upend our understanding of criminal activity.
Dao, an assistant professor of geography at UCCS, developed CrimeScape with her former mentor, Jean-Claude Thill of the University of North Carolina at Charlotte.
CrimeScape begins with the association rule mining — seeking patterns or other connections in databases — often used in criminology research, and adds a geospatial element to analyze how certain crimes take place over time and neighborhoods. The first practical application developed by Dao and Thill looks at car theft — but its implications could reach far beyond law enforcement. Dao sees the use of ARM-based artificial intelligence as immediately applicable to problems in health care and environmental land-use, with other applications limited only by the imagination.
ARM is based on unsupervised machine learning, a branch of AI that has exploded in the last decade through its reliance on artificial neural networks. Unsupervised networks can examine large data sets and come to inferences and conclusions on their own, ones that often escape the attention of humans. Traditional geographic information systems and geolocation systems were not often linked to such local/regional tools, in part because they grew out of large-area space-based GIS tools, and big software companies like ESRI. Dao’s work suggests an opportunity for analysts to use GIS software tools that operate from the bottom up. In her own CrimeScape work, Dao has discovered distributions of car thefts in urban areas that bring to light counterintuitive patterns.
Dao grew up in Vietnam, with an early interest in working with geolocation tools like the Global Positioning System. She performed early undergraduate work at the University of New South Wales in Australia, then received a master of science degree in satellite-based navigation from the University of Calgary in Alberta. She then followed her marriage partner to the United States, where she completed her Ph.D. in geography and regional urban analysis at UNC Charlotte. From 2013 to 2016, she was an assistant professor at the University of Montana, then took her current UCCS post in 2016.
Dao talked to the Business Journal as she prepared to return to Vietnam for the first time since before the pandemic.
Did you come to Colorado specifically for UCCS’ expertise in geospatial information?
The CU system at large, including CU Boulder and CU Denver, was known for geospatial expertise. The job in Colorado Springs came open at just the right time, and I was very much attracted by the news of 300 days of sunshine a year in this area. Even in 2016, Colorado Springs was ranked as one of the top places in the nation to live. Now when we first arrived, the summer was very rainy, but then the promise of sunshine came, and I was very attracted to the city and the campus. The weather itself has been changing year to year, of course, but my kids love camping, love the mountain scene.
How have you found the GIS field changing? For a long time, the field has been dominated by big companies and military contractors.
The field has been evolving a lot recently; it is much more open to new ideas. If you look 20 or 30 years ago, it was all GPS data for surveying large areas. Nowadays, GIS is integrated into so many fields, it plays a role in physical and social sciences of many types. Now there is just so much information, everything is recorded from parcels to business records to neighborhood information, hiking trails and transportation information, health records, you name it. And when everything has a digital data component, the problem becomes one of managing and analyzing the data. By adding the geospatial component, you can see how entities change over time. Once, GIS was a static look at how things are laid out in space; now you see how things evolve in time. The big companies in GIS were engineering-based; now it is more integrated science.
Were you familiar with unsupervised machine learning before beginning the CrimeScape project?
A lot of my work has been backed up by geography and by regional urban analysis, the topic of my degree at Charlotte. I realized we were dealing with data that had extra components. Why not apply some of those emerging robust computer science methods? Computer scientists can show that they can handle large data sets fast, but they do have the limitation of not always being able to explain the data they found. Maybe this is because we need more domain experts in the areas where machine learning is being applied.
A lot of the initial CrimeScape idea was the result of work on my dissertation. I took a few computer courses and became generally aware of associative rule mining, which had mostly been used in transaction-based applications such as online shopping carts. My advisor, a professor in public policy, supported me in extending this to new applications.
It became clear that I should focus more on relationships among various entities, and the interactions among them, rather than purely focusing on location information. Association rules have been a very robust way of mining within a complex system. This can be true whether we are looking at crime, climate change, conservation ethics, sustainability, transportation systems. The interactivity of components in complex systems is very hard to model.
Looking at CrimeScape, it was interesting to see that your work could override some pre-existing intuitive assumptions we may have, such as where car thefts take place, and the socio-economic environment involved.
That can be important in the way inductive and deductive reasoning plays into predictive studies. This framework could be useful in reducing institutional biases in many fields.
It seems that CrimeScape was a pilot model, not just in criminology, but in exploring how broadly your analytical methods could be applied.
A lot of people said, ‘Oh, you are focusing on criminology now!’ I am not actually focused on that, but on a much broader area. One of the reasons you have not seen the geographical field use many robust techniques so far is that geography traditionally has been hosted in the social sciences, and we have been working with a poor data environment for a long time. Over the past 10 years, geography finally opened up to the era of big data.
We have begun the work to expand the framework to other areas — to public health, for example. We have defined a new framework to identify how humans carry on with daily activities, including mobility, under various conditions of health. (The GEMA framework project has been carried out by UCCS and University of Montana.) This uses GPS information on iPhones collected every 5 minutes, corresponding to information on physical and mental health. It involves self-reporting on how individuals are feeling, who they are with, and we try to associate these patterns with community participation.
We are in another study to examine the special characteristics of human-caused wildfires in western Colorado, and how human behavior is changing over the years. We have looked at large data sets for human-caused wildfires from 2002 to 2018. We are relating this information to transportation changes, particularly post-2000 and completion of part of the interstate road system.
How difficult is it to talk to people in vertical application domains, like land use or behavioral health, about what this framework can really do?
I can tell you that we sought to have the CrimeScape paper published in Social Science Research, rather than a general-purpose GIS journal, so that the implications were understood in other domains. The editor for that journal is trying to broaden the audience for this kind of work. Suddenly I feel I am playing in a new type of playground. But of course, I learned a lot, I learned from the domain experts of criminology. When I looked at my work under the lens of criminology, I discovered new types of complexity. The learning takes place in both directions. It has been challenging and rewarding at the same time. If you want to be useful in GIS, you must understand its use in many realms, and apply it in many fields.
It sounds like there could be a lot of work done with undergraduates to explore the integration of GIS into all these different vertical fields.
There are significant movements across the country and worldwide to better merge GIS and computer science. You often are able to get a combined degree in GIS and CS. What is taking place already for the undergrad is very promising for the future. New combinations of geo-visualization using big data, and statistical visualization from the data mining field, are arising.
Do you see yourself affiliated with UCCS for a long time to come, both for research in this framework and for teaching?
The university is unique in bringing in older students, veterans, and those who might be more focused in particular disciplines than the average undergraduate student body. And Colorado Springs has many companies working in GIS, so students in the field have a good chance of finding work after graduation. I find this is a wonderful school and wonderful city in which to be pursuing this work.