Project 01

Building permit cost predictor

Construction stakeholders in Calgary need reliable cost estimates early in the planning process. This project uses 484K+ historical building permits and machine learning to predict project costs from permit characteristics, location, and scope.

XGBoost Regression Feature engineering Calgary open data 484K permits

0.89 R-squared (XGBoost)

Interactive app

Streamlit dashboard pages

Overview

Dataset summary, distributions, and key statistics across 484K building permits

Exploratory analysis

Cost distributions, community comparisons, and temporal trends with interactive charts

Model performance

Compare Ridge, Random Forest, Gradient Boosting, and XGBoost on key regression metrics

Feature importance

Community-level aggregates and permit characteristics driving cost predictions

Performance

Key results

0.89

R-squared

XGBoost regressor

~$30K

Mean absolute error

On log-transformed costs

484K+

Permits analyzed

Calgary open data API

Approach

Methodology

Fetched 484,000+ permits from the Calgary Open Data API via Socrata. Cleaned and log-transformed the heavily right-skewed cost distribution. Engineered community-level aggregates (average and median cost, permit counts) plus temporal features. Trained and compared Ridge, Random Forest, Gradient Boosting, and XGBoost regressors, selecting the best performer by R-squared and MAE.

01 Fetch 484K permits via Socrata API

02 Clean and log-transform costs

03 Engineer community aggregates

04 Train four regression models

05 Evaluate and select XGBoost

06 Deploy Streamlit dashboard

Resources