Laura Camila Guayacan Ardila · LinkedIn
One question, three approaches: what drives construction cost, and which projects deviate from what we would expect?
The same dataset of 6,123 construction projects is analysed with regression, SQL and an interactive dashboard, using the vocabulary a project manager uses to make decisions.
- Equipment is a largely fixed cost (about $100k per project): 31.7% of total cost on projects under 2,000 sq ft versus 7.2% above 10,000 sq ft.
- Cost per sq ft outliers fall from 423 to 205 once equipment is removed, so most of the apparent deviation is a scale effect.
- Weather and risk indices do not explain the outliers.
| # | Project | Tools | Skills shown |
|---|---|---|---|
| 1 | Cost driver analysis | Google Sheets | Multiple regression, cost estimation, hold-out validation |
| 2 | Cost variance and outlier analysis | SQL (BigQuery) | Variance analysis, benchmarking, outlier detection |
| 3 | Project cost dashboard | HTML, JavaScript | Cost reporting, what-if estimate with range, plain-language interpretation |
index.html Landing page
dashboard/ Interactive cost dashboard
projects/cost-drivers/ Case study 1: regression (README + charts)
projects/sql-outliers/ Case study 2: SQL (README, queries.sql + charts)
- Regression results were cross-checked in Python (statsmodels): all coefficients match to 4 decimals.
- SQL results were cross-checked against the regression workbook.
- In the dashboard, the "AI" layer only interprets results with fixed, rule-based text; it does not calculate.
Source: Kaggle "Deep Learning Building Cost Dataset". It is public and likely synthetic, so findings illustrate method rather than real-world cost benchmarks. Each case study lists its own limitations.