Machine Learning in Construction: An Introduction (part 1 of 2)
Michael GogginDirector, Enstoa
Machine learning is an exciting concept in the construction industry. To us at Enstoa, where we specialize in finding meaning in construction project data, machine learning helps improve processes and project outcomes.
By applying machine learning techniques to the project record, we can improve our understanding of performance trends, identify areas of opportunity for continuous improvement initiatives, and make better decisions faster by leveraging our data as a decision support system.
Drowning in small data
If there is one thing that every company who has participated in a major capital development project understands, it is that the project management systems generate high volumes of small, yet meaningful, data.
Small data tends to be well structured and packed with meaning; data from project controls, finance, contract and asset management, and other information systems can be combined to provide a rich dataset, even more so when we look back and include historical data or audit logs and change histories.
As data sources proliferate, individual data points become increasingly available and increasingly connected. Extracting meaning out of complex and related webs of data is precisely where machine learning excels.
What is machine learning, really?
Machine learning, itself a branch of the field of artificial intelligence, does not have a single, convenient definition.
Arthur Samuel, a pioneer in AI, offered the following, informal definition in 1959: "[machine learning is] the field of study that gives computers the ability to learn without being explicitly programmed.” During Samuel’s time, in the mid-20th century, an enormous amount of research into artificial intelligence techniques was being conducted. Theories were developed, algorithms were created – there simply wasn’t enough computing power to handle any kind of rich, industry data. Machine learning wasn’t yet ‘scalable’.
Computers like this one in the 50s had less calculating power than today's portable calculators.
Today, modern processors, distributive computational resources, reams of rich data, and a little know-how have made machine learning a convenient tool that generates insights in real time. This is leading to an explosion of development in analytics as data scientists set out to employ cutting-edge machine learning techniques to generate meaning from data.
The many faces of machine learning
Machine learning can take many forms: from the simple (e.g., a standard linear regression model) to the complex (e.g., a neural network ‘deep learning’ implementation that mimics human decision-making), from the small (e.g., a clustering analysis using project management data, as will be presented below) to the large (e.g., a distributed file system that processes real-time CCTV footage to identify and eliminate safety hazards as they are occurring). The field of machine learning is both broad and deep.
Figure 1 – Different methods of machine learning are applied to different types of problems.
The difference between supervised and unsupervised learning
Broadly speaking, machine learning can be divided into two general categories: supervised and unsupervised learning.
In supervised learning, we have an output or response variable we care about (for example, a cost performance metric). Techniques are applied to the data to either predict or better understand the drivers of the response variable. No matter the model or method, in supervised learning the response variable is the star of the show.
Unsupervised learning, on the other hand, relies crucially on having no such response variable. In such cases, we have no objective variable we want to describe or outcome we want to predict, we simply want to explore the data at our disposal. Unsupervised methods are often employed at the preliminary or data exploration stages.
The benefits of machine learning in construction and beyond
A recent analysis by Deloitte found that companies that invest in machine learning capabilities see early and increasing economic benefits that scale with the number of deployments - benefits such as better decision making and improved utilization of human resources.
At Enstoa, we understand that machine learning can drive process improvement. Equipping teams with a meaningful analysis of small data allow them to hone their workflow, assign talent to the projects that best suit their strengths, and avoid pitfalls that have historically derailed project timelines and/or budgets.
