IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

Using Time Series Forecasting to Transform Capital Projects

Infrastructure pivots on complex, long-term planning involving millions of dollars. But with modern data methods, argues Balaji Sreenivasan, the government can achieve more confidence about what the future holds.

David Kidd/Governing
In the ever-evolving landscape of capital infrastructure projects, government agencies find themselves performing an intricate dance. The heightened focus on the timely and budget-conforming completion of infrastructure assets — be it roads, bridges or transport systems — has intensified the need for precise and transparent project management. The challenges are particularly acute given the intense scrutiny that these projects now face, coupled with ever-expanding investments.

Too often, unforeseen external factors are blamed when projects fail to meet stakeholder expectations. This phenomenon, known as the “planning fallacy,” stems from the human inclination to lean toward optimistic outcomes rather than adopting a more balanced perspective of potential gains, risks and outcomes. Recognizing this bias, the tide is turning on how governments approach capital project planning. Deloitte's research suggests that by 2024, over 60 percent of government AI and data analytics investments will directly influence real-time decision-making.

Time series forecasting has indeed revolutionized the way we envision infrastructure development. In this article, we will explore how AI-driven time series forecasting can be harnessed by capital project handlers for better economic predicting and decision-making.


Time series forecasting is not just a buzzword but a tangible tool. Using sophisticated AI models such as ARIMA (autoregressive integrated moving average), SARIMA (seasonal ARIMA) and LSTM (long short-term memory), infrastructure owners and capital planners can delve deep into future demands, patterns, and even run intricate "what if" simulations. Essentially, it allows us to tap into historical data, decode patterns and predict future occurrences, effectively addressing the pitfalls of the “planning fallacy.”

However, predicting through time series comes with its challenges. Factors like economic shifts, policy amendments, workforce dynamics, community sentiment, supply chain perturbations and environmental influences are continuously evolving. These variables, often interlinked and fluctuating, necessitate the advanced capabilities of time series forecasting and its foundational AI models.

In simple terms: ARIMA evaluates past events to forecast the future, SARIMA considers recurring patterns such as seasonal demand fluctuations and LSTM operates like a data conveyor belt, identifying long-term trends. By integrating these methodologies, government entities can harness more tailored, accurate and strategic forecasting.


Consider New York City's ambitious 10-Year Capital Strategy (TYCS). This $165 billion initiative aims to rejuvenate diverse infrastructure elements, all while centering human perspectives. The unique approach of TYCS emphasizes continuous collaboration, ensuring community sentiments are given weightage. This plethora of data is ripe for time series forecasting, evolving the insights over time. By processing this data, agencies can predict various impacts — from demographic shifts to environmental changes — influencing a project's long-term value.

Massachusetts offers another apt illustration. The Department of Transportation meticulously identifies key factors like traffic volume, environmental considerations and infrastructure needs when strategizing transportation networks. Many of these factors, given their dynamic nature, are prime candidates for time series analysis.

On a broader canvas, even federal agencies such as the Federal Highway Administration delve into multifaceted variables when conceptualizing infrastructure. In a recent policy report, the agency outlined a string of key variables that range from societal demographics to economic indicators, all of which play a crucial role in determining future infrastructure demands.


Acknowledging the multifaceted variables in capital projects is one thing; effectively engaging with them is another ball game. While traditional forecasting methods have their merits, they are susceptible to human biases and errors. Time series forecasting, bolstered by models such as ARIMA, SARIMA and LSTM, ensures that decisions are made based on robust data analytics rather than mere chance.

As we integrate these machine learning models, government agencies are better positioned to maximize investment returns, enhance economic foresight and refine decision-making processes. Grounding decisions in past data and holistic event understanding, governments can create a ripple effect of positive, enduring impacts for their communities, ushering in a brighter and more resilient future.

Balaji Sreenivasan is founder and CEO of Aurigo Software Technologies.