In this interview with Guardian’s Racheal Olatayo, Nigerian US-based Civil Engineer Joyce Lewis discusses her innovative Decision Support System (DSS), which leverages machine learning, specifically a Random Forest model and Monte Carlo simulations to enhance the accuracy of construction cost estimations.

What inspired you to use machine learning for construction cost estimation instead of traditional methods?
Traditional cost estimation methods often rely on manual calculations, historical data, and expert judgment, which can lead to errors and budget overruns. I wanted to find a more accurate and efficient way to predict construction costs. Machine learning can analyze large amounts of data quickly and detect patterns that humans might miss. By using it, we can make better cost predictions and reduce financial risks in construction projects.
[ad]
Can you explain how the Random Forest model was selected over other machine learning algorithms?
I tested different machine learning models, but Random Forest stood out because it is more accurate and reliable for predicting house prices. It works by combining multiple decision trees, which helps reduce errors and improve prediction accuracy. Also, it handles missing or inconsistent data well, making it a great choice for real-world construction data.

How does the Monte Carlo simulation contribute to improving cost prediction accuracy?
Construction costs are unpredictable due to factors like material price changes, labor costs, and market fluctuations. Monte Carlo simulation helps by running thousands of possible cost scenarios and giving us a range of possible prices instead of just one estimate. This makes the predictions more realistic and helps construction managers plan for uncertainties.

Among the key features analyzed, why do you think living area, zip code, and year built had the most influence on cost estimation?
Living area (square footage): The bigger the house, the higher the cost. More materials and labor are needed.
Zipcode: Location affects prices. Houses in high-demand areas cost more than those in less popular neighborhoods.
Year built: Older houses may need more renovations, while newer houses are often built with updated materials and technology, affecting overall cost.

What challenges did you encounter while developing and validating the Decision Support System (DSS)?
One major challenge was data quality. Some housing records had missing or incorrect values, which affected predictions. Another challenge was making the model general enough to work in different locations, since construction costs vary from place to place. Lastly, I had to ensure that the GUI (Graphical User Interface) was simple and easy for construction professionals to use, even if they have no background in machine learning.

How does the DSS GUI enhance user experience for stakeholders involved in construction planning?
The GUI is designed to be simple and interactive. Users just need to enter details like the number of bedrooms, bathrooms, and square footage, and the system instantly predicts the cost. This helps construction managers and developers make faster and more informed budgeting decisions without needing complex spreadsheets or manual calculations.

The model achieved an R² score of 0.70—how do you interpret this in terms of predictive reliability, and what improvements could be made?
An R² score of 0.70 means the model is fairly accurate but not perfect. It explains 70% of the variation in housing prices, which is good but leaves room for improvement. To make it better, we could add more factors like interest rates, inflation, and labor costs to improve accuracy. Also, using more advanced AI techniques could further enhance prediction quality.

How could external economic factors, such as inflation and material costs, be integrated to further refine cost predictions?
Right now, the model focuses mostly on property features. However, construction costs are also affected by things like inflation, rising material prices, and supply chain disruptions. By pulling in real-time market data on these factors, the model can adjust predictions based on economic conditions, making it even more useful for developers.

In what ways do you see Geospatial Information Systems (GIS) transforming cost estimation in the future?
GIS can map out construction trends and risks based on location. For example, it can show how housing prices change across different neighborhoods or how factors like traffic, flood zones, and infrastructure development impact costs. In the future, combining GIS with machine learning could make cost predictions even more precise and help planners make better location-based decisions.

What are the next steps in expanding this model, particularly in applying it to rental price predictions?
Rental prices are influenced by many factors, including demand, location, and economic trends. The next step is to train the model using rental market data so it can predict rental costs just like it does for housing prices. This could help landlords set fair rents and help tenants find affordable housing. Additionally, I want to make the model adaptable to different cities, so it can be useful in different real estate markets.

Can this model be adopted for other areas like transportation infrastructure projects?
Yes, absolutely! The same machine learning principles can be used to predict costs for transportation infrastructure projects, such as road construction, bridges, and rail systems. Instead of using house-related factors like square footage and number of bedrooms, we would input data like road length, materials used, labor costs, and location to estimate costs. This could help government agencies and private contractors plan infrastructure projects more efficiently, reduce cost overruns, and allocate resources better. In fact, combining this with GIS would make it even more powerful for transportation planning.
[ad]