Early Emergency Detection and Project Risk Prediction for NYCHA
Written by Mohammadreza Ahmadnejadsaein (Reza)
As a PiTech PhD Impact Fellow this summer, I worked on two data-driven projects with the New York City Housing Authority (NYCHA), both aimed at improving the safety and resilience of their large portfolio of buildings. NYCHA manages hundreds of developments across the city, many of which face aging infrastructure and growing maintenance challenges. The goal of my projects was to explore how machine learning could help NYCHA identify certain risks earlier so that resources can be directed where they are needed most.
The Challenge
My projects addressed two challenges that NYCHA is currently facing. First, structural issues in buildings are often detected too late, when emergencies already demand urgent repairs. This puts residents at risk while also straining limited budgets. Second, capital projects, such as major renovations, frequently encounter costly delays caused by coordination problems, funding gaps, or contractor setbacks. Both issues share a common thread: although NYCHA has some historical data, predicting risk from this data is difficult. This is because the available data is limited, messy, sometimes biased, and often fails to capture the full picture of NYCHA’s portfolio.
Innovative Approach
To address each of these issues, I created structured datasets and machine learning models. For structural emergency detection, I tested a variety of models, including logistic regression, support vector machines, decision trees, random forests, XGBoost, and neural networks. These models translated building characteristics such as age, history of flooding, cracks, and deficiencies into risk scores that could help inspectors prioritize site visits. For project delay prediction, I created a comprehensive dataset tracking every project phase from planning through construction, combining details on daily progress, budget commitments, and vendor information. I then used a machine learning model to predict trends in project schedule delays and commitment changes. The analysis highlighted the key factors contributing to delays and uncovered non-linear relationships between project characteristics and project outcomes. This would give construction managers a tool to see how individual project factors could influence project results in complex ways, challenging the traditional assumption that changes in a factor always affect risk in a single, predictable direction.
Impact and Future Decisions
Although the limited size of the data meant these models were not ready to serve as final decision-makers, they showed strong potential as decision-support tools. Inspectors, for example, could now start with a list of buildings ranked by risk level rather than relying on ad hoc scheduling of inspections. Construction project managers could spot early warning signs of delays, allowing for earlier intervention. With more labeled data and refinement into phase-specific or project-level predictors, the model could become even more accurate and actionable.
Mohammadreza Ahmadnejadsaein (Reza)
Ph.D. student, Operations Research and Information Engineering, Cornell University
NYCHA plans to continue using the models and methodologies I developed this summer to identify high-risk buildings for further inspection. They are also building on the initial models for project cost and schedule analysis, which could help improve performance and resource allocation within their limited budget.
By embedding machine learning into its operations, NYCHA’s is preparing for a future where scarce resources are deployed more efficiently, residents are better protected from emergencies, and project timelines are managed with greater foresight.