2021 Capstone Projects
Did you miss our 2021 Capstone Showcase and Recognition Ceremony? Learn more about the event here:
- Center for Urban Science and Progress gives its new grads their moment to shine (via NYU Tandon Newsroom)
View the videos for:
(NYU Login required)
Politecnico di Milano Students: Davide Calabrò, Stefano Fedeli and Marco Valli
A Finance Map of NYC: Reducing Carbon and Driving Large Scale Energy Efficiency with a Public Database To Support PACE Lending
CUSP Students: Haiqiang An, Huanran Kong, Yuanzhe Luo, and Yuetong Zhou
Faculty Mentors: Gary Friedland, Marianna Koval, and Miles Draycott
Project Website: https://yl7989.wixsite.com/website
NYU Stern Center for Sustainable Business is supporting NYC’s commitment to limit carbon emissions under the Paris Agreement, collaborating closely with the NYC Mayor’s Office of Sustainability (MOS) and the NYC Clean Energy Efficiency Corp. (NYCEEC) to bring low-cost, long-term financing (PACE) for energy efficiency retrofits and renewable energy to buildings. To achieve NYC carbon neutrality by 2050, Invest NYC SDG, a two-year Stern CSB initiative, wants to guide priorities and outreach to building owners and lenders by creating a public database with visual interface mapping to present GHG emissions, potential penalties, and open liens of almost 40,000 buildings. The CUSP Capstone project would help Invest NYC SDG Initiative to develop an interactive map of NYC that displays CO2e emission level by building, contact details for the building owner or managing agent and a list of lien holders and the results of a series of queries — either run in advance or possibly user defined.
Predictive Modeling of Opioid Overdose Risk for Targeted Public Health Interventions
Sponsor: Machine Learning for Good Laboratory
CUSP Students: Jiaqi Dong, Nicholas Liu-Sontag, Brandon Pachuca, and Yicong Wang
Faculty Mentor: Daniel Neill
The opioid epidemic is one of the largest public health crises in the United States; since 1999, over 814,000 people have died from a drug overdose in the US. Rhode Island has been hit particularly hard and regularly has some of the country’s highest overdose death rates. To improve forward-looking targeting of intervention efforts they seek to utilize a prediction model of areas of the state at higher risk of an overdose outbreak. As a subset of a larger team working on this effort, we developed four models to predict overdose risk at the census block group level utilizing the following algorithms: gaussian processes, random forest, gradient boost, and graph convolutional network. The first three of these achieved the project’s baseline performance target and the graph convolutional network we believe shows promise. Our model results will be folded into the larger project and this information will then be supplied to RIDOH and community organizations to deploy targeted resources to higher-risk areas. If this method proves successful, it could serve as a model for states and municipalities across the country to identify and target interventions to reduce overdose risk.
Modality-Agnostic Road Traffic Monitoring
CUSP Students: Eve Shi and Yao Hou
Faculty Mentor: Magdalena Fuentes and Bea Steers
State-of-the-art machine learning models of traffic monitoring focus on single-modal methods using only video, video-image or audio. Even though they achieve high performance in vehicle recognition and vehicle density classification, single-modal analysis is limited by illumination conditions and noisy interferences. Previous research shows promising results using multi-modal solutions compared to their single-modal counterparts, but this hasn’t been explored in depth yet in traffic monitoring. In this project, we present a two-step multi-modal framework which works with both audio and video-images. In particular, we use pre-processed video-image and audio embeddings and project them in a common space to obtain cross-modal embeddings, then train modality-agnostic classifiers with these exchangeable embeddings. We also assess the performance of the classifiers with varying amounts of data from each modality, s. The results of the multi-modality framework show compatible accuracy in vehicle counts classification with single-modal solutions, and it can fit different types of data thus being more flexible. Furthermore,, we curated and annotated a dataset of audio-visual traffic data provided by Bosch and Tampere University, which includes videos under different weather and environment conditions, being the first-of-its-kind audio-visual dataset with bounding boxes annotations and audio.
Do Broken Windows Encourage Criminality?
Sponsor: Dynamical Systems Laboratory
CUSP Students: Catherine Liu and Mengyuan Guo
Faculty Advisor: Maurizio Porfiri
Mentors: Alain Boldini, Emma Schechter, and Roni Barak Ventura
Project Website: https://brokenwindows2021.wixsite.com/website
Since the broken window theory was put forward in 1982, it has influenced many cities’ policing strategies. However, in recent years, there have been more and more criticisms that believe that the decline of the crime rate is affected by many factors, which can not prove that broken window policing is effective. And so far, no literature can verify the authenticity of this theory through data. To answer this question, our study mainly uses 311 requests data and NYPD complaint data of New York City from 2015 to 2019 to research the correlation relationship between disorder behavior and crime on both time aspect and spatial aspect. According to the research results, we put forward suggestions to improve the efficiency of the police.
Mapping Construction+Demolition Waste Flows for Recovered C+DW Use in City’s Capital Program
Sponsor: Town+Gown:NYC and CDW Working Group
CUSP Students: JP McKay, Parth Singhal, and Dina Wagdy
Faculty Mentor: Eric Corbett
Project Website: https://accomplishedcode.github.io/NYU-CUSP-Capstone-Mapping-CDW/
Our project visualizes the flow of construction and demolition waste (CDW) generated in the Long Island and New York City regions in order to provide context for policymaking that encourages local CDW recycling and reuse. The project included the transformation of multiple sources of regulatory reporting into a structured format, the amalgamation of the sources into a single machine-readable dataset, and the production of a spatial visualization tool to allow interactive inquiry of the dataset by nontechnical users. The tool provides policymakers and industry stakeholders a means of spatially examining annual flow trends in aggregate by material types, transactions, and end use (i.e., disposal, recycling, or designation as beneficial use). The dataset behind the tool enables other data professionals to perform analysis on historical CDW flows and thus better plan for the sourcing of locally produced CDW as an ecologically friendly material resource.
Open Police Data: Collection, Analysis, and Outcomes
Sponsor: Results for America
CUSP Students: Anna Carlson, Wanyu Cao, and Kendra Singh
Faculty Mentor: John Pamplin
Project Website: https://openpolicedata.netlify.app/
Our sponsor, Results for America, is interested in how widely available open police data are because they are looking to potentially consider police data transparency as a factor in their city certification process. Results for America helps cities benchmark their progress toward a well-managed, data-driven government. By providing them with information on the availability of open police data and the five key types of police data (use of force, stop and frisk, vehicle stops, police demographics, and officer complaints), Results for America will be able to elevate the importance of police transparency in open data standards for cities across the country. We will address the research questions: How widely available are the five essential types of open police datasets? What percentage of major U.S. cities make these data openly available? The outcomes of our project are an internal report for Results for America and an external website for researchers to access information on open police data for major U.S. cities. There are limited amounts of data available on open data availability and granularity, meaning this project and the external website specifically serve as a valuable resource for researchers, the general public, and policymakers when evaluating police transparency across the US.
Sponsor: Urban Modeling Group
CUSP Students: Gabriella Barrett and Sean Andrew Chen
Faculty Mentor: Debra Laefer
Project Website: https://urbaninventories.github.io/ui.github.io/
As LiDAR technology becomes cheaper and more readily accessible, point cloud data of urban areas in tandem becomes not only more abundant but also of higher fidelity and resolution. This data can be of great use to city officials and planners for many purposes, including the inventorying and cataloging of objects within the urban and built environments – objects such as buildings, street furniture, traffic lights, and lampposts. To accomplish this inventorying, data scientists can utilize state of the art as well as conventional machine and deep learning techniques for urban object detection, segmentation, and classification. There are, however, more than a multitude of methods – each with specific advantages and disadvantages. This project will compile a database of these methods and analyze which are most adept at the detection, segmentation, and classification of urban objects using metrics tailored for the purpose of urban inventorying.
Evaluating the Geometric Properties of 2D Urban Layouts Using a Data-Driven, Deep Learning Approach
Sponsor: AI4CE Lab
CUSP Students: Sheung Chan, Lazarus Chok, and Yan Liang Tan
Project Website: https://xaviercslung.github.io/citygeometry/
Urban planners widely use geometric properties to understand cities, but current methods of quantifying city geometry are very limited. To objectively measure, compare and evaluate urban designs, urban planners need a standard metric to quantify the geometric properties of 2D urban layouts. We use a proprietary tree-structured neural network developed by the AI4CE Lab to encode the complex geometry of urban layouts and learn an expressive representation of city geometry. Using the latent representation of urban layouts, we propose three urban planning applications. Firstly, we classified Manhattan’s urban layouts using GMM clustering algorithms and Shapley interpretations to identify eleven distinct urban layout typologies. Secondly, we derived a standard metric for describing the urban layout composition of neighborhoods using dimensionality reduction. Thirdly, we generated composite urban layouts by decoding the linearly interpolated values of two distinct latent subspaces. These proof-of-concept applications validate our hypothesis that complex city geometry can be encoded using representation learning techniques for downstream urban planning applications.
California Dream Index
Sponsor: California Forward (CA FWD)
CUSP Students: Michael Carper and Aren Kabarajian
Faculty Mentor: Eric Corbett
Project Website: https://ark510.wixsite.com/ca-dream-capstone
The California Dream Index (CDI) was started by the nonprofit organization California Forward to provide policymakers and other stakeholders with an understanding of socioeconomic mobility around the state. The CDI currently stands as a web tool displaying regional and county-level statistics on ten indicators of economic mobility. At county granularity, the tool lacks the precision necessary to track the economic livelihood of specific neighborhoods and smaller areas. The capstone team addressed this issue by re-creating the CDI’s interactive state map visualizations at census tract granularity. The completion of this goal will provide local level stakeholders with the power of tangibility when lobbying to improve areas of economic vulnerability. In a second phase of the project, the capstone team performed a series of California resident clustering analyses using some of the same indicators scored in the map visualizations as features. The first of these, a comprehensive analysis of residents in 2019, revealed groups of economic archetypes based on eight of the CDI indicators. The second clustering exercise was a series of analyses performed separately on each decade from 1990 to 2019, relying only on housing, education and commute times as features. The team hypothesized that cluster scores on the selected indicators of economic mobility would all correlate positively with cluster incomes. While some of the selected features proved to be strong, consistent indicators of economic status, others displayed more nuanced relationships with income than expected.
Predicting and Preventing Lead Paint Poisoning in New York City Among Young Children
CUSP Students: Tyler Matteo and Zhiwei Fan
Faculty Mentor: John Richard Pamplin
Project Website: https://funfan2020.github.io/
Lead paint remains the leading cause of lead paint poisoning in children nearly half a century after its use was banned nationwide. Inhalation of lead paint dust can have devastating lifelong effects on the development of young children. Efforts to identify and remediate lead paint hazards in NYC have been further complicated by opaque landlord registration data as well as the COVID-19 pandemic. In partnership with the Office of the NY Attorney General, our project aims to identify and model trends in lead paint exposure, compile a list of repeat offender landlords, and analyze what effects the pandemic has had on mediation efforts. These insights will be key as the city continues to address these hazards through targeted outreach and intervention.
Applications of Machine Learning and Remote Sensing for Assessing Open Street Map Completeness on a Global Scale
Sponsor: New Light Technologies, Inc
CUSP Students: Alba Alsina Maqueda, Medwin Chiu, and Wenpeng Zhao
Faculty Mentor: Luis Alfredo Ceferino Rojas
Project Website: https://osmcompleteness.cargo.site
In the context of extreme growth and urbanization, climate change has accelerated and accentuated the risks of natural disasters and constant migratory movements alter territorial balances. The need for complete and accurate mapping, particularly those provided by the platform, OpenStreetMaps (OSM), is essential in ensuring the resilience of territories. However, the availability and capacity of access to cartographic data vary within and among countries, with large gaps in coverage and quality in many parts of the world. This report summarizes the work done by a team at New York University’s Center for Urban Science and Progress (NYU CUSP) that modifies a model created by New Light Technologies (NLT) for the World Bank in assessing OSM completeness in road networks, through satellite data-based classification algorithms. Three areas of interests (AOIs) that varied in size and risk were selected as the study for this project. This script became a “proof of concept” for a single tool that will allow stakeholders to either determine OSM completeness, train datasets, or improve the modelling capabilities of OSM roadways at critical AOIs around the world, and aid in the improvement of resilience and economic growth of these areas.
New York State’s Broadband Coverage Mapping for the Reimagine New York Commission
CUSP Students: Kelsey Nanan, Aleka Raju, and AJ Kuhn
Faculty Mentor: Qi Sun
Project Website: https://nysbroadband.wixsite.com/nys-broadband
The Reimagine New York Commission has ratified an initiative to establish an accurate baseline of consumer broadband experience in New York State. This initiative is meant to position New York as a national leader in broadband data collection, provide information for use by any community for broadband expansion projects, and support pursuit of grant funding available from the FCC, USDA, and future state programs. Through this capstone, the CUSP team aggregated multiple datasets at various spatial levels to create a broadband coverage map for New York state. This map highlights a broadband score, meant to determine accessibility of broadband connections as well as demographic factors across the census tracts. In addition to a comparison view, through which users are able to see speed data from different sources like the FCC, Ookla, etc.