The challenges: navigating complexity and data limitations

When we embarked on this research journey, our goal was clear: to improve the accuracy of predicting construction project durations by integrating external economic and municipal factors. However, the path was fraught with challenges.

One of the most significant hurdles was the lack of consensus in existing literature about which factors truly influence construction delays. While some studies highlighted economic variables like inflation or exchange rates, others dismissed them as negligible. This inconsistency made it difficult to justify our focus on external factors, especially when traditional models prioritized project-specific variables like cost and area.

Another challenge was data accessibility. We relied on a dataset of 372 construction projects in Tehran spanning 1993–2011, but recent data was unavailable due to bureaucratic and logistical barriers. This limitation forced us to work with older data, raising concerns about relevance to today’s dynamic economic climate. Additionally, the dataset lacked granular details, such as project-specific disruptions (e.g., COVID-19, labor shortages), which are critical in modern construction contexts.

Technical hurdles also emerged. Implementing the CatBoost ensemble model required extensive tuning and validation, and integrating it with neural networks (ANN) demanded meticulous coding to ensure compatibility. The sensitivity analysis (SA) phase was particularly painstaking—each input variable had to be iteratively removed and tested to assess its impact, a process that consumed weeks of computational time.

The successes: breakthroughs and validation

Despite these challenges, the project yielded transformative insights. Our hybrid ensemble model (Stacking-CatBoost2) outperformed traditional single and ensemble methods, reducing prediction errors by 29 days compared to ANN and 2.5 days compared to standalone CatBoost. This validated our hypothesis that combining models could leverage their individual strengths—CatBoost’s handling of categorical economic variables and ANN’s adaptability to nonlinear patterns.

The sensitivity analysis revealed surprises. While "lot area" (a project-specific variable) had the strongest positive impact, external factors like "price of gold per ounce" and "consumer price index" ranked second and third. This underscored the often-overlooked role of macroeconomic stability in construction timelines. Conversely, "the number of loans extended" had a negative impact, suggesting that excessive financing might correlate with inefficiencies or mismanagement.

A pivotal moment was validating our model against real-world benchmarks. When compared to three prior studies, our Stacking-CatBoost2 achieved at least 5% higher accuracy and a 0.148 better correlation coefficient. This wasn’t just a statistical win—it demonstrated the practical value of our approach for contractors and policymakers.

Implications for future research

Our work opens several avenues for further exploration:

Dynamic economic factors: Future studies should incorporate real-time economic data (e.g., supply chain disruptions, geopolitical events) to enhance model responsiveness.

Global generalizability: Testing our framework in diverse regions could reveal how local policies or economic structures influence outcomes.

Integration with BIM/IoT: Pairing our model with Building Information Modeling (BIM) or sensor data from construction sites could refine predictions at the micro-level.

Human factors: Qualitative research is needed to explore how managerial decisions interact with our quantitative findings.

The untold stories: collaboration and perseverance

Behind the scenes, this project was a testament to teamwork. Early in the process, a coding error in Python caused the CatBoost model to misclassify key variables. It took days of collaborative debugging—and countless cups of coffee—to trace the issue to a misplaced normalization step. Moments like these highlighted the importance of interdisciplinary collaboration; our team’s blend of civil engineering, data science, and economics expertise was critical to overcoming obstacles.

Another untold story is the resistance we faced from traditionalists. During peer reviews, some questioned the need for complex AI models when "simple regression has worked for decades." This pushback reinforced our commitment to demonstrating the tangible benefits of machine learning—not as a buzzword, but as a tool for precision.

A personal reflection: why this matters?

As a researcher, I’ve always been drawn to the intersection of infrastructure and economics. Growing up in a city where delayed projects were the norm, I saw firsthand how budget overruns and missed deadlines eroded public trust. This project was more than an academic exercise; it was a step toward empowering stakeholders with data-driven foresight.

One anecdote stands out: A contractor in Tehran, upon reviewing our preliminary findings, remarked, "If we’d known the gold price would spike, we’d have stocked materials earlier." This casual comment encapsulated our mission—to turn hindsight into foresight.

Visualizing the journey

To bring this story to life, we’ve included two key visuals:

Figure 2 (SA results): A ranked bar chart showing the surprising influence of gold prices and CPI on delays ([see original paper]).

Figure 8 (Correlation charts): Heatmaps illustrating how economic variables interlink more strongly than project-specific ones.

Sensitivity analysis results

“Figure 2: External economic factors (e.g., gold prices) outweighed many project-specific variables in impacting delays.”

Final thoughts

This research was a reminder that construction isn’t just about concrete and cranes—it’s a mirror of societal and economic forces. By bridging data science and civil engineering, we’ve shown that even "uncontrollable" factors can be measured and managed. To fellow researchers: embrace the messiness of real-world data. To practitioners: let evidence, not intuition, guide your timelines. The road ahead is long, but with collaborative innovation, we can build a future where delays are the exception, not the norm.