James Martin

Scannable QR code with contact info

Scan contact card

Researcher and data analyst using big data, machine learning, and statistical methods. I consult with organizations to help minimize risk through prevention science and actualize their strategy through data flows and endpoints. Always seeking opportunities to improve operations and learn new skills along the way. If my abilities can be of value to your organization or project let’s get in touch.

Skills

General
Languges
Visualization
Systems, Web
GIS

American Retail Crime, Shrink, and Security (ARCSS) Initiative

The American Retail Crime, Shrink, & Security Initiative is an ongoing effort dedicated to collecting, analyzing, and reporting industry trends in loss and violence within stores. Data is gathered from participating retailers and standardized for consistency. Summary statistics are generated by zip code and retail sector. Eventually, ARCSS aims to enhance the insights obtained from the National Retail Security Survey, conducted annually in collaboration with the National Retail Federation (NRF). This initiative will equip policymakers with the necessary data to shape legislation aimed at safeguarding both workers and customers.

Read more..

Research2Practice Articles

Research2Practice articles cover trending topics that also tie-in to larger projects being perused at the LPRC. The aim is to discuss approaches for addressing key concerns within the industry using analytics and knowledge from practitioners in the field. These articles also serve to institutionalize learning, creating an ongoing record of the most pertinent challenges faced by loss prevention and asset protection professionals.

Read more..

Data Analytics Working Group

The LPRC’s Data Analytics Working Group (DAWG) enables retailers, manufacturers, and technology companies to exchange perspectives and recent developments in the utilization of data analytics for enhancing safety, minimizing theft, and optimizing operations. In addition, this working group includes demonstrations designed to instruct both managers and analysts on innovative approaches to better understand their data. These hour-long demonstrations cover a wide range of topics.

Read more..

Crime Risk by ZIP Code

Ziprisk is a simple web app that allows users to easily view socio-demographic factors that correlate to crime risk at the ZIP code level. The data is taken from the 2021 American Community Survey conducted by the U.S. Census Bureau which is publicly available using the census API. On the back end this is a Python Flask application using SQLAlchemy for data retrieval.

Read more..

LPRC SaferPlaces Dashboard

In collaboration with the Gainesville Police Department (GPD), University of Florida Police Department (UFPD), and Ersi inc., the Loss Prevention Research Council (LPRC) created an up-to-date crime incident map for the city to launch the SaferPlaces research initiative. Calls for service data are piped into ArcGIS online and visualized as a heat map along with LPRC StoreLab locations and field assets.

Read more..

Organized Retail Crime Across the States (ORCATS) Survey

The Loss Prevention Research Council (LPRC), Loss Prevention Foundation (LPF), and Loss Prevention Magazine (LPM) conducted an industry-wide survey of investigators and multi-store loss prevention managers to understand the causes, consequences, and control of Organized Retail Crime (ORC) across the United States. I processed the survey response data, generated summary statistics, and developed a dashboard to communicate findings.

Read more..

Inverse Distance Weighting

A common geostatstical technique used to interpolate values across a study area based on a number of explicit sample locations. I had trouble finding a package that implemented this technique in a simple way which proved to be a problem when using IDW to interpolate air pollution concentrations at millions of locations. To make it computationally feasible I coded IDW and an accuracy metric as c++ functions and included them in a package.

Read more..

Health Accessibility Models

Health accessibility models have been used to identify regions that are medically under-served due to their population and distance from medical facilities. I used these models to identify neighborhoods in the city of Huaquillas that may lack access to vector control services. I also simulate how different placements of workers and resources could improve access and identify locations to optimally place a new facility.

Read more..

Spatial Networks

There are many powerful R packages that allow users to perform network analyses. However there currently isn’t the built-in functionality to use these analyses on spatial datasets. To study a road network for example, you need to convert the shapefile GIS format into an object that is accepted by a package like ‘igraph’. I provide an example on how to perform these conversions and use network analysis to find the shortest route between two locations in the road network.

Read more..

Publications

Academic papers, preprints, and talks

Paper - Raising the Watchtower: Results from the ACCESS Task Force

The Alliance of Companies and Communities to Enhance Safety and Security (ACCESS) task force was established by LiveView Technologies to streamline research and development for their line of mobile surveillance units. The Loss Prevention Research Council (LPRC) collaborated with retailers and law enforcement in two cities to investigate the effectiveness of surveillance towers in reducing various types of crime. We utilized a pre vs. post quasi-experimental study design to quantify changes in calls for service around the installed units. The study revealed that the placement of units was associated with a reduction in certain types of property crime, and these associations were further confirmed using spatial regression models. Additionally, we distributed a survey to community members to assess changes in their perceptions of safety.

October 2023

Paper - Bayesian profile regression to study the ecologic associations of correlated environmental exposures with excess mortality risk during the first year of the Covid-19 epidemic in lombardy, Italy

We used a semi-parametric Bayesian modeling framework to characterize social and environmental covariates to COVID-19 related mortality in the Lombardy region of Italy. Bayesian profile regression creates clusters similar to unsupervised machine learning methods while also enabling statistical inference. We identified complex relationships between environmental pollutants and demographics suggesting that policy proposals must be holistic and spatially specific in order to substantially reduce COVID-19 mortality. As co-author my role involved compiling environmental data, modeling ambient pollution, and validating model output.

January 2023

Talk - The Big Picture: The Role of Mapping in Retail Crime Analysis and Risk Estimation

For many years Geographic Information Systems (GIS) have been used within retail make vital decisions from selecting sites for new stores to planning transportation routes to optimize supply chains. GIS is now being applied to the aims of loss prevention. In this presentation at the LPRC’s 2022 IMPACT conference, I introduced GIS and its applications to LP managers and practitioners by demonstrating how spatial statistics can enhance investigations and assist retailers in selecting locations for enhanced protection following multiple victimization events.

October 2022

Paper - Household and climate factors influence Aedes aegypti presence in the arid city of Huaquillas, Ecuador

The Aedes aegypti mosquito is a vector for Dengue fever, Chickgunya, and other viral diseases which are now endemic in Ecuador. This species thrives in high density urban environments and is highly adaptable. To understand the social and ecological factors that bring together humans and mosquitoes, we conducted a survey of households in Huaquillas regarding demographics, living conditions, and prevention strategies. Mosquitos were also collected from these households periodically using backpack aspirators. Using ordinal logistic models we uncovered important risk factors for households with Aedes aegypti mosquitoes and a beta regression helped us gauge the influence of climatic factors on the population dynamics of the vector.

November 2021

Paper - A time series analysis of the ecologic relationship between acute and intermediate PM2.5 exposure duration on neonatal intensive care unit admissions in Florida

We paired over one million birth records in the state of Florida with daily air pollution estimations and other demographic data to explore the influence of air pollution on neonatal intensive care unit (NICU) admissions from 2012 to 2019 at the zip code level. Elevated air pollution was associated with an increase in admissions when accounting for confounding factors. As co-author my role involved acquiring and processing environmental data and estimating air quality using multiple geo-statistical methods.

May 2021

Talk - UF MathGeo Model: A Mechanistic Approach Using Vector Traits and Georeferenced Climate Data

This mechanistic model was made for a competition held by the CDC’s Epidemic Prediction Initiative to forecast the probability of detecting two species of disease transmitting mosquito in a number of counties a month in advance. Our model combined data from laboratory studies on the effects of temperature on life traits as well as weather forecasts to predict presence. Our predictions were then validated against actual catch data from mosquito control agencies in participating counties.

February 2020

Experience

Research Scientist

I work with major retailers and solution providers to conduct basic and applied research in retail loss prevention. I advise retailers on how best to collect and analyze their data to improve processes. I also facilitate networking and information sharing among professionals in the loss prevention community. Some types of projects I’ve worked on include shrink modeling using location attributes, store level risk profiling, brand sentiment analysis utilizing social media, pre-post outcome analysis for deterrents, and forecasting of key performance indicators.

2022 - present

Research Assistant

My main tasks involved acquiring and processing large environmental datasets, often drawing upon my GIS abilities. I implemented machine learning models using ensemble methods to forecast environmental hazards like ambient air pollution. Projects I worked on include investigating the effects of air pollution on neonatal intensive care unit admissions, exploring the social and environmental dimensions of COVID-19 related mortality in Northern Italy, and characterizing the occupational health challenges of gulf coast seafood workers.

2020 - 2022

Graduate Research Assistant

I carried out a social-ecological study of Aedes aegypti risk in southern coastal Ecuador which required extensive data cleaning and analysis. I used hierarchical models within a multi-model, information criterion approach to investigate risk factors. Other projects I worked also concerned Ecuador and arboviral disease. These include identifying spatial-temporal clusters of Dengue fever using point pattern analyses and using health access models to simulate alternative vector control regiments.

2017 - 2019

Undergraduate Research Assistant

I was involved in a project which aimed to forecast the future range of African Trypanosomiasis under future climate conditions using a mechanistic model based on the biology of the Tsetse fly vector. I searched the literature for laboratory studies to inform model parameters and also wrote code to process these data and fit non-linear functions to them. The Quantitative Dynamics Lab was originally based at USF before moving to Virginia Tech.

2015 - 2017

Education

University of Florida

Master of Science, Geography

2017 - 2019

University of South Florida

Bachelor of Science, Public Health

2015 - 2017